In traditional AI infrastructure, pricing is rigid and upside is centralized. As usage grows, costs scale linearly, and the value created by developers and power users is largely captured by intermediaries. This model works for hyperscalers, but it breaks down for AI-native products, domain-specific models, and teams that rely heavily on inference as a core input.

FOMO flips this structure by enforcing a simple principle: value only flows when real usage happens. Models are launched through a Real Model Asset Offering, where users and RMAs supporters commit capital to fund a concrete deployment. Tokens are not minted upfront. Only if the launch meets its funding threshold does the model graduate, tokenize, and proceed to deployment. If it doesn’t, the launch simply doesn’t exist.

Find FOMO on fomo.flock.io.

Once deployed on the FLock API Platform, the model begins serving real inference. From that point on, usage becomes the only signal that matters. Inference revenue is routed programmatically to buy back the model’s token, buy back the network token, fund ongoing operations, and reward the model operator. Early incentives help bootstrap adoption, but they decay over time as organic usage takes over. If a model isn’t used, rewards naturally disappear.

This design is intentional. FOMO is not a speculative launchpad, and it does not attempt to manufacture demand through emissions alone. Each model deployment becomes its own small, transparent economy, where incentives align around adoption rather than narrative. Power users are rewarded for driving real demand, operators are incentivized to maintain competitive pricing and quality, and the protocol remains sustainable without propping up idle capital.

FOMO also completes the loop with AI Arena, FLock’s decentralized training platform. AI Arena incentivizes the production of intelligence, while FOMO incentivizes its dissemination and use. Together with the API Platform, they form a full-cycle DeAI system where training, deployment, and consumption reinforce one another.

This is just the beginning. As more models are launched through FOMO and more inference flows through the API Platform, we’ll continue refining the mechanisms that turn real usage into aligned economic outcomes.

If you’re building with AI, operating models, or scaling inference-heavy applications, FOMO is built for you. Stay tuned as it rolls out across the FLock ecosystem.

Step-by-step guide

This guide walks through the FOMO launch lifecycle from both sides:

• Real Model Asset (RMA) Issuers launching a model

• Users and RMAs Supporters participating in a model’s fundraising

For RMA Issuers

Step 1: Select a Model

1. Go to the Launch page.

2. Select your desired model.

Each model corresponds to a predefined inference pricing tier and an allowed fundraising range. These parameters are fixed by the platform to ensure economic sustainability.

Step 2: Configure Your Model Token ($MT)

Configure $MT economics:

1. Name your deployment and define:

   • Model token name

   • Ticker symbol

   • Description

2. Review the pre-configured parameters, including:

   • Inference pricing tier

   • Fundraise min / max range

   • Token distribution summary

These values are automatically populated based on the selected model and cannot be arbitrarily changed.

Step 3: Bundle the First Buy Order (Optional)

RMA issuers may choose to bundle an initial token purchase at launch.

• This provides a first-mover advantage by locking in the lowest point on the bonding curve

• The first buy is executed atomically with token deployment, removing market-timing risk

The bundled first buy is subject to a hard cap of 25% of the total $MT supply to prevent excessive concentration at launch.

Note: The bonding curve applies throughout the entire fundraising period, dynamically pricing tokens as demand evolves — not just at launch.

Step 4: Confirm Launch Fee

Before initiating the Real Model Asset Offering (RMO), the RMA issuers must confirm and approve the launch-related fees.

This confirmation is required to proceed with model deployment and on-chain configuration.

• One-time launch fee. A fixed, non-recurring fee of $32768 FLOCK payable by the RMA issuers at the time of initiating the RMO. This fee covers deployment setup, onchain configuration, and protocol overhead associated with launching a new model economy.

• Combined tax (base fee plus anti-sniping tax) starts at 99% and decreases by 1% per minute, until it reaches the base fee.

Step 5: Launch the Model Token

Once confirmed:

• The RMO is created

• The fundraising period 7 days begins

• Tokens are not minted during fundraising period

You can now find your deployment on the Marketplace page.

Status indicators:

• Owned — launched successfully by you

• Live — fundraising in progress

• Failed - fundraising target not met and pledged principal refunded

Step 6: Fundraising Period

• Fundraising remains open for 7 days

• Progress toward the minimum and maximum targets is tracked in real time (please refer to “Fundraise Outcomes” section for details)

• No action is required during this period

For Users and RMAs Supporters

Step 1: Discover a Model Token

1. Navigate to the Marketplace page.

2. Browse available model deployments and select a model token you want to support.

Step 2: Participate in the Fundraising

• Swap $FLOCK → $MT via the bonding curve

• Token price increases as more $FLOCK flows in

• A 1.5% platform transaction fee applies to each swap

Note: $MT purchased during fundraising cannot be sold back into the internal market during fundraising period.

Step 3: Wait for Fundraise Outcome

• The fundraising remains open for a fixed period.

• Progress toward the minimum and maximum funding targets is tracked in real time.

• No further action is required until the fundraising concludes.

Fundraise Outcomes

Scenario 1: Happy Path — Maximum Target Reached

• Graduation is triggered automatically

• Tokens are minted

• Liquidity is created

• The model proceeds to deployment

Scenario 2: Happy Path — Minimum Target Reached

• Graduation must be triggered manually

• Graduation is permissionless that anyone can trigger it

Action:

• Click “Graduate Token” to launch the liquidity pool

Scenario 3: Unhappy Path — Minimum Target Not Reached

• The fundraise fails

• Tokens are not minted

Action:

• Click “Redeem” to reclaim fundraised $FLOCK

• Launch fees and transaction fees are not refundable. 

• Only the pledged $FLOCK principal is returned.

After Graduation

Once graduated:

• The model token becomes tradable

• Marketplace status updates to “Tradable on Deluthium”

• Clicking the token redirects users to the Deluthium swap page

• Note: Deluthium is currently not available in the US

From this point forward:

• The model will be deployed on the FLock API Platform

• Inference usage drives revenue, buybacks, and reward

Find the Chinese version here / 中文版本

Tokenomics is often where good ideas go to die.

Too many protocols rely on emissions without demand, or “utility tokens” without real cash flow. FOMO was designed from the opposite direction: start with real AI usage, then build token mechanics around it.

This post explains how FOMO’s tokenomics work — using a concrete example — and why they are fundamentally different from speculative launchpads or inflation-driven incentive systems.

Two Tokens, Two Roles

FOMO operates with two distinct token layers, each with a clear purpose.

$FLOCK — the Macro Token

$FLOCK is the network-wide asset of the FLock ecosystem. Its role is to:

• Coordinate incentives across all deployments

• Serve as the reward and governance token

• Capture value at the protocol level

Crucially, $FLOCK is deflationary by design:

• A portion of inference revenue across all models is used to buy back $FLOCK.

• Emissions are fixed and routed only to productive deployments, not idle capital.

$FLOCK represents aggregate demand for AI intelligence across the network.

Model Tokens (MT) — the Micro Assets

Each deployment launched through FOMO has its own Model Token (MT).

An MT represents:

• A specific model

• A specific deployment configuration

• A specific inference economy

Each MT has:

• A fixed supply (1,000,000 tokens)

• A transparent allocation (sale, liquidity, incentives, operator, treasury)

• Direct exposure to revenue-based buybacks and burns

In short:

$FLOCK captures network-level value; MT captures model-level value.

How a Model Becomes a Real Asset

Let’s make this concrete with an example.

Alice launches QwenALICE

Alice is a Real Model Asset (RMA) Owner. She wants to deploy a high-quality model on the FLock API Platform.

She launches QwenALICE via a Real Model Asset Offering (RMO).

• Total MT supply: 1,000,000 QwenALICE

• 40% sold in a fair launch

• Users buy using $FLOCK

• Funds are used for:

  • Protocol costs

  • Staking incentives

  • Liquidity provisioning

If the raise fails, everyone is refunded.
If it succeeds, the model goes live and starts serving inference.

Already, this is different from most launchpads:

• No token without a real model

• No model without real hosting

• No liquidity without locked alignment

  

Usage Comes First, Tokens React

Once QwenALICE is live, users pay for inference in USDC or fiat equivalents, through our very own FLock API platform - it’s basically the same how API works in other platforms like OpenAI - we intend to keep fiction of onboarding extremely small. 

From that point on, usage is the only thing that matters.

Each dollar of net inference revenue (after compute costs) is split deterministically:

• 30% → Buy back $FLOCK

• 30% → Treasury (operations, sustainability)

• 30% → Buy back and burn QwenALICE

• 10% → Paid to Alice as RMA yield

This means:

• $FLOCK becomes scarcer as any model is used

• QwenALICE becomes scarcer as this specific model is used

• Alice earns real cash flow, not just tokens

There is no speculation required for value to accrue — inference alone is enough.

Where Emissions Fit (and Why They Don’t Break the System)

FOMO does use emissions — but very intentionally.

Demand-Driven Emissions

Each day, a fixed amount of $FLOCK is emitted.
These emissions are distributed only to models that are being used.

The allocation depends on a Deployment Score:

• More revenue → higher score

• Newer deployments → higher weight (via an age factor)

This achieves two things:

1. Rewards models that users actually want

2. Prevents emissions from being captured forever by incumbents

   

Internal Emission Split

When QwenALICE receives its share of daily emissions:

• 10% goes to Alice (RMA bonus)

• 90% goes to stakers, weighted by gross usage

This is where Bob enters the picture.

Bob: Why Stakers and Users Are the Same Person

Bob is a power user of QwenALICE.

During the RMO:

• He buys QwenALICE tokens early

After launch:

• He stakes them

• He gets discounted inference

• He receives $FLOCK rewards

Bob’s incentives stack:

1. Lower inference costs via staking

2. $FLOCK emissions proportional to his usage

3. Exposure to MT deflation as QwenALICE tokens are burned

Importantly, rewards are calculated using gross spend, not discounted spend.

So Bob is not penalized for using discounts:

• He pays less

• But still earns rewards as if he paid full price

This is a critical design choice — it aligns power users instead of punishing them.

Why This Doesn’t Collapse Under Inflation

Most token systems fail because:

• Emissions grow faster than real demand

• Rewards are disconnected from usage

• Tokens exist without cash flow

FOMO avoids this by enforcing four constraints:

1. Fixed MT supply
   No infinite minting.

2. Revenue-funded burns
   $MT are burned via real usage.

3. Capped discounts
   Inference never goes “free” at the protocol level.

4. Emission routing based on revenue
   No usage = no rewards.

As a result:

Emissions bootstrap adoption,
but consumption ultimately dominates supply.

Tokenomics as a Franchise Model

The deeper insight behind FOMO is that model deployments behave like franchises.

• Alice is a franchise operator

• Bob is both a customer and a stakeholder

• The protocol is the franchisor

Early participants:

• Accept risk

• Receive upside via emissions and appreciation

Later participants:

• Enjoy lower prices

• Benefit from mature liquidity and stability

This is not possible in traditional cloud pricing.
It is only possible because tokens allow capital formation to subsidize usage.

Why This Matters for the AI Market

FOMO introduces a new baseline for AI inference economics:

• Models are priced competitively because users own part of the upside

• Distribution is rewarded, not taxed

• Incremental model upgrades must justify their cost
  
  Over time, this removes artificial pricing power from centralized AI APIs and forces real efficiency and innovation.

Final Thought: Tokenomics That Follow Reality

FOMO’s tokenomics are not designed to “pump” tokens.

They are designed to do one thing well:

Route real AI usage into aligned economic outcomes.

If a model is used:

• Tokens burn

• Rewards flow

• Operators earn

• Users save money

If it isn’t:

• Emissions fade

• Tokens stagnate

• Capital moves elsewhere

That is what honest tokenomics look like.

And that’s what FOMO is built to enforce.

Find the Chinese guide here / 中文版本

FLock.io’s fair launchpad is finally here! The launch of FLock Open Model Offering (FOMO) finalises FLock’s evolution into a full-cycle AI platform with a balanced supply-demand ecosystem.

As a unified interface, FOMO streamlines access to models on our API Platform and provides discounted hosting for AI inference. We welcome all models to fair launch on FOMO – not just ones trained on AI Arena – and get visibility on platform.flock.io. 

Celebrating one year of mainnet operations, FLock has built a robust network of high-quality models and engineers. As the world’s leading decentralised AI (DeAI) infrastructure, we’ve been busy building a complete flywheel: AI Arena, FL Alliance, API Platform and native token $FLOCK bring the supply, and now FOMO brings the demand.

Read our other blog to learn more about how our franchise model leverages the token economy to disrupt pricing – solving the cost floor problem that hampers traditional cloud and gateway providers. 

It’s all been leading up to this launch, and we couldn’t be more proud. Read on to find out more about getting started on FOMO.

Who should launch their model tokens on FOMO?

Anyone is welcome to launch model tokens in exchange for hosting services on the FLock API Platform – not just models trained on our flagship training platform AI Arena (train.flock.io). 

If you have strong sales capabilities, the franchise model will play in your favour due to $MT appreciation. You can sell the model hosted to your network. Any company that uses AI on a daily basis will also benefit from joining FOMO for its guaranteed cost efficiency. 

Step-by-step: FOMO launchpad and API revenue flow

Users currently relying on major cloud or gateway providers for AI model APIs can use FOMO to initiate a fair launch for their model needs in exchange for hosting services. Following a successful launch, the model will be hosted on the API Platform, allowing for seamless integration and usage at a discount.

This is a high level step-by-step of the FOMO launchpad and API revenue flow:

1) Launch Initiation

Real Model Asset (RMA) issuers and supporters officially initiate the issuance of Real Model Assets on the FOMO Launchpad by raising $FLOCK.

2) Hosting & Tokenisation

Upon successful issuance, the model will be hosted on the FLock API Platform for external access. Simultaneously, the corresponding Model Tokens ($MT) will be minted.

3) Revenue Allocation

Revenue generated from model usage will be distributed across four channels:

1.  $MT buyback and burn.

2.   $FLOCK buyback.

3.  RMA revenue sharing.

4. Payment for future model hosting costs.Token Utility. 

Users who stake $MT enjoy two key benefits:

1.  Usage Discounts: Discounts on model calls (calculated based on a function of the user's staked amount to the total circulating supply).

2.  Staking Rewards: Earning additional $MT through staking.

   4) Early Incentives

During the initial launch phase of a model, the platform will distribute additional $FLOCK as rewards to users staking that specific model's $MT, based on actual usage volume.

Example case study: how launching a model on FOMO works

To illustrate the financial mechanics of the FOMO launchpad and the resulting value accrual, let’s walk through a scenario: User Alice launches a high-utility model, “QwenALICE”.

Phase I: The Successful RMO Launch

Alice initiates her offering, successfully raising the predetermined amount in $FLOCK and selling all sale tokens. This capital is immediately deployed: a portion funds the initial DEX Liquidity Pool, ensuring the $QwenALICE asset is tradable and liquid post-launch, while the rest secures operational runway and staker rewards. The launch is risk-mitigated, as funds are only secured if the minimum goal is met.

Phase II: The Deflationary Revenue Waterfall (Month 1)

In its first month, QwenALICE generates Net Revenue (profit after compute costs). This profit immediately drives value through the Revenue Waterfall:

• Model Buyback & Burn: 

A portion of the Net Revenue is used to buy $QwenALICE from the open market and permanently burn it. This ensures utility (usage) directly translates into asset appreciation via reduced supply for token holders.

• Protocol $FLOCK Buyback: 

A portion is used to buy the macro $FLOCK token and these $FLOCK tokens will be utilised for ecosystem growth and further incentivisation, ensuring model utility fuels the deflation of the entire network's core asset.

• RMA Operational Yield:

A portion is distributed to Alice as direct cash flow, incentivising her to drive further adoption.

Value Captured by Alice (The Franchisee)

Alice’s investment delivers three streams of value. Namely cash flow (operational yield), asset appreciation (her vested tokens rise in value due to the buyback-and-burn mechanism) and token yield (she receives a large share of the daily $FLOCK emissions, which are weighted based on the model's high revenue and favorable Age Factor). 

This proves the model delivers predictable financial returns driven by real-world AI consumption.

More about FLock.io

Find out more about FLock.io by reading our docs and blogs. See a sneak peak of new launches coming in Q4 2025 and our FLock 2025 earnings report: performance & growth overview. Stake, train and earn on train.flock.io.

For future updates, follow FLock.io on X.

Find the Chinese guide here / 中文版本

The monumental launch of FLock Open Model Offering (FOMO) finalises FLock’s evolution into a full-cycle AI platform with a balanced supply-demand ecosystem that drives innovation across the industry. Find it on fomo.flock.io.

FLock has been busy building a system that makes the creation of specialised, high-value AI models a financially sustainable and profitable activity, without requiring the central control of a major centralised corporation. This is the positive sum game of decentralised AI that everyone benefits from.

As the world’s leading DeAI infrastructure, we have a complete flywheel. Now that the supply side is strong from AI Arena, FL Alliance, AI Marketplace, API Platform and native token $FLOCK, it’s time for FOMO to bring the demand.

As a one-of-a-kind unified interface, FOMO streamlines access to models on our API Platform and provides discounted hosting for AI inference. Its power lies in its disruptive pricing. By leveraging the token economy, its franchise model uniquely solves the pricing problem that hampers traditional cloud or gateway providers.

Celebrating one year of mainnet operations, FLock has built a robust network of high-quality models and engineers. We have seen exceptional growth and forged major partnerships with the likes of Alibaba Qwen and the NHS and became the United Nations Development Programme (UNDP)’s AI strategic partner.

We welcome all models to fair launch on FOMO – not just ones trained on AI Arena – and get visibility on platform.flock.io. If you’re a developer or AI company looking to get started with FOMO, see our other blog to find out how.

FOMO finalises FLock as a full-cycle AI platform with balanced supply and demand

The finalisation of our ecosystem is a leap forward for the decentralised AI movement.  FOMO is the final, demand-driven piece that completes the world’s first full-cycle DeAI platform. We’ve been busy building for the past two years, and the supply-side parts of our ecosystem have excelled. 

AI Arena is our flagship platform for competitive and collaborative model training. It demonstrates how traditional centralised AI development is far from the only way to build powerful models. Our native on-chain token $FLOCK enables tokenomics mechanisms designed to enhance and incentivise secure ecosystem participation.

FL Alliance is a privacy-focused collaboration framework that enhances models while preserving data sovereignty – a critical feature for enterprise and government partners like the NHS. 

Through AI Arena and FL Alliance, we’ve seen the practical potential for FLock technology to have a tangible impact in numerous corners of life and the world. These parts of our ecosystem bring the supply, and FOMO is the way AI companies will find and use models.

The problem with existing cloud providers & gateways: sky-high fees

Until FOMO, the standard prices for AI inference were set by cloud providers (like AWS and Alibaba Cloud) and gateways (like OpenRouter and SiliconFlow). Their prices are dictated by the cost of compute, since they cannot price inference below hosting costs without making a loss.

Traditional cloud providers typically offer AI inference through two primary models that anchor costs to raw compute: usage-based token pricing and provisioned infrastructure fees. Since they bear the full, non-subsidised cost of the hardware, a cloud provider can never price the model inference below the raw compute and operational costs without incurring a direct loss.

Traditional gateways are excellent for streamlining the use of AI models. They transform the fragmented and costly world of AI into a manageable, secure, and cost-optimised utility. But they still rely on purchasing compute from the underlying cloud providers or model owners. They are forced to operate by applying a small markup or passing through the base cost (token pricing) established by the cloud provider. They cannot lower the cost floor.

Breaking traditional pricing floors for AI inference with a franchise business model 

The API Platform is the gateway, and FOMO is the fair launchpad. Together, they provide developers with hosting and seamless access to models from various AI providers through a single API. 

Unlike other gateways and launchpads, FLock adopts a “franchise” business model, which is only possible through a token economy. FOMO achieves dramatically lower prices than current market rates by incentivising token holders to subsidise inference costs while profiting from appreciated model tokens. The selling point is twofold: AI companies and builders get cheaper inference, and model token holders benefit from increased token value the more the model gets used.

The platform provides the infrastructure (kitchen), the Real Model Asset (RMA) issuer acts as the franchise owner (distributor), and the $FLOCK token incentivises the bootstrap phase. This design ensures that incentives are high during the risky early phase and taper off as the model achieves organic market fit.

FLock provides the underlying platform and infrastructure to host models. RMA issuers initiate offerings and self-use or solicit model users to boost the model revenue. The $FLOCK token incentivises bootstrap phase, capturing value via Protocol Fee Buybacks.

FOMO is taking on traditional centralised AI

FOMO fundamentally changes the competitive dynamics and pricing power of major centralised AI providers (like OpenAI, Google, Anthropic). Beyond correcting the cost structure, FOMO acts as a crucial market regulator that addresses the anticompetitive practices of centralised AI labs.

Centralised labs rely on an incremental release strategy. They release marginally better model versions (e.g. GPT-4.1, 4.2) and price them based on their proprietary cost structure and their already established market dominance (“legacy benchmarks”). They don't have to deliver revolutionary leaps because enterprise customers, once integrated, face high switching costs and have no viable low-cost alternatives. This allows them to maintain a “valuation bubble” with prices that exceed the true cost-to-value ratio.

FOMO introduces a new, ultra-low, competitive baseline for inference that is mathematically unachievable by centralised providers. Unless a new centralised model offers cost-efficiency gains that vastly outweigh FOMO’s ultra-low inference rates, enterprises will lack the financial incentive to incur switching costs for a marginally better “update.”

In doing so, FOMO forces centralised labs to deliver genuine innovation (not just incremental improvements) to justify their premium pricing. This dynamic serves as a necessary market check against monopolistic pricing power, securing the long-term, equitable viability of the DeAI movement.

FLock’s token economy keeps prices low and benefits model token holders

The core constraint of all traditional cloud providers and gateways is the price floor: they cannot charge less than their hosting costs. FLock’s FOMO model breaks this barrier by turning the constraint into an investment opportunity – token issuance enables inference to be priced below long-term equilibrium margins in the bootstrap phase by shifting early-stage margin risk.

The tokenomics is tried and tested. We’ve been enhancing our mechanisms all year to enhance ecosystem participation and designed the tokenomics to be sustainable and secure. FOMO lets anyone launch model tokens in exchange for hosting services on the FLock API Platform. With this launch, specific model tokens will be issued. Users can stake them to secure discounted API rates.

Unlike speculative launchpads, FOMO anchors assets to real utility. Each model token represents genuine demand for model usage. To combat inflation of the total supply, a portion of all revenue generated by the API Platform will be used to buy back and burn the relevant model tokens. As usage increases, the buy-back-and-burn mechanism reduces supply, driving value accrual for the designated model tokens and allows users to benefit from asset appreciation.

The system prioritises consumption over emission. $FLOCK is utilised across various participation mechanisms within the FOMO launchpad, establishing a positive economic flywheel where ecosystem demand outpaces token supply.

No centralised authority is necessary to enforce the deflation. Instead, the entire trustless economy is managed by transparent smart contracts with autonomous governance. 

An incentivisation strategy underpins FOMO’s tokenomics, allowing AI inference rates to undercut traditional pricing. Model token holders subsidise the margin for more utilisation of the models – that is, they forfeit some margin but can in the long run earn more by model token ($MT) deflation. The RMA issuer’s ultimate incentive is not the profit margin earned from a single inference transaction, but the appreciation of their $MT assets.

During the bootstrap phase only, there will be $FLOCK emissions. During this critical period, emissions will incentivise genuine adoption. Specifically, $FLOCK rewards will be distributed to those staking on actively used models. Users get significantly lower inference costs through upfront discounts via model token staking and retroactive rebates via $FLOCK emissions.

More about FLock.io

It’s all been leading up to the FOMO launch, and we couldn’t be more proud. Find out more about FLock.io by reading our docs and blogs. See a sneak peak of new launches coming in Q4 2025 and our FLock 2025 earnings report: performance & growth overview. Stake, train and earn on train.flock.io.

For future updates, follow FLock.io on X.

Find the Chinese guide here / 中文版本

FLock.io, Base’s largest AI infrastructure protocol, is hosting the hackathon in collaboration with Base and Alibaba Cloud’s Qwen. We invite students, developers, content creators and whoever is interested in developing Base and AI applications.

The demo day will be held in Seoul, South Korea, with online attendance also possible.

It's an opportunity to support app builders and share with the community to invigorate Base's diverse application ecosystem. While the previous Qwen SKYST Hackathon was a datathon focused on creating AI models, this hackathon involves building actual user-ready applications utilising those models in the form of Base Mini Apps.

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

There are two tracks. Both use the FLock API Platform, which is similar to the OpenAI API, and build a Base Mini App.

1. Hackathon Track 1: Build an AI Application utilizing FLock.io’s API Platform as a Base Mini App.

2. Hackathon Track 2: Build an AI Application/Agent utilising FLock.io’s API Platform and x402.

The Kickoff Session will have a vibe coding tutorial on both building a Base Mini App and utilising the FLock.io API Platform. See a Vibe Master demo video and the website.

Projects will be judged on the following criteria: application quality (30%), FLock API utilisation (30%), novelty (20%) and practicality (20%). Teams will need to produce a 5-min presentation and demo of the application (video), presentation slides (PDF) and projection details. 

Prizes of up to $5,000

Together, we are allocating $15k USD in prizes with teams of 1-3 people having a shot of winning up to $5,000.

1st place: $5,000

2nd place: $3,000

3rd place: $2,000

Finalists (4-8): $1,000

Timeline from sign-up to demo day

Registration: 11.13 ~ 11.22

API Distribution: 11.21 ~ 11.23

Kickoff: 11.22 13:00~15:00 UTC+9

• Online (It may be distributed in the form of a pre-recorded video)

Online Hackathon: 11.22~11.28 18:00 UTC+9

Submission Deadline 11.28 18:00 UTC+9

• The hackathon submission platform & communication tool will be announced

Top 8 finalists announcement: 11.28 23:59 UTC+9

Finalists Demo Day: 11.29 14:00~18:00 UTC+9

Demo day for hackathon finalists, 29th Nov 2025

Finalists will be invited to attend the demo day at Dreamplus Gangnam B1, Seoul, South Korea on 29th November, 14:00~18:00 UTC+9.

Timetable

14:00~14:30 Registration

14:30~14:45 Closing remarks by FLock.io (Jiahao Sun, CEO)

14:45~15:00 Closing remarks by Base (David, Head of East Asia)

15:00~15:15 Closing remarks by Alibaba Cloud (Stone, Head of Business Development)

15:15~16:30 Top 8 Team Presentation

16:30~17:00 Break Time & Top Team Judging

17:00~17:15 Top Finalist Announcement & Award Ceremony

17:15~18:00 Networking

More about FLock.io

FLock.io is the first decentralized AI training platform combining Federated Learning and blockchain to enable secure, privacy-preserving model training. It empowers communities to collaboratively build and own AI without centralizing data. Powered by the $FLOCK token, FLock drives governance, incentives, and ecosystem growth. Backed by top investors and the Ethereum Foundation, it’s reshaping AI’s future.

Find out more about FLock.io by reading our docs and blogs. See a sneak peak of new launches coming in Q4 2025.

For future updates, follow FLock.io on X.

More about Qwen

Qwen is a family of large language models developed by Alibaba Cloud. Built to understand and generate both Chinese and English, Qwen is designed for a wide range of applications including natural language processing, reasoning, and multimodal tasks. As an open and scalable AI foundation, Qwen powers chatbots, coding assistants, and enterprise solutions across Alibaba’s ecosystem and beyond.

Follow Qwen on X.

More about Base

Base is an Ethereum Layer-2 blockchain network developed by Coinbase. It is designed to improve scalability, performance, and interoperability while reducing transaction costs and settlement time on the Ethereum network. As a decentralized platform, Base supports a wide range of dApps and smart contracts.

Follow Base on X.

More about SKYST

SKYST is a consortium of computer clubs from four schools: Seoul National University, Korea University, Yonsei University, KAIST. From 2023, they have been organizing annual hackathons of 100~150 participants, with the purpose of revitalizing the student developer ecosystem and strengthening student developers’ engineering capabilities through collaboration between the four schools.

AIKU

AIKU is a student-led academic society at Korea University where students interested in studying deep learning come together to engage in education, research, applied projects, and discussions. AIKU was founded in the summer of 2022 by students from the Korea University School of Informatics, under the guidance of Professors Kim Byung-woo and Kim Seung-yong from the Department of Computer Science. Today, it includes students from diverse academic backgrounds.

KUCC 

Korea University Computer Club (KUCC) is the oldest and largest computer club within Korea University, being founded in 1973, through various activities related to computers, KUCC focuses on improving the technical capabilities of individuals and exposing them to various experiences with computers.

YCC

Yonsei Computer Club (YCC) was founded in 1970 and is the only centralized computer club at Yonsei University. At YCC, students actively participate in full-stack project contests/competitions to gain project experience as it is not only composed of developers but designers and project managers.

SNUAI 

Seoul National University AI Club (SNUAI) is both a university based club and an incorporated startup with an investment of an initial valuation of $250K. Being the only club within Seoul National University that deals with AI in a practical way, they focus on researching and commercializing AI through their own corporation.

Waffle Studio

Waffle Studio is a web/app service development club based in Seoul National University, School of Computer Science, where it aims to build a community of developers within the department. Founded in 2007 and established as the largest development club on campus, WaffleStudio holds various activities related to development such as projects, studies, and coding meetings. Through projects, they produce development results and contribute to people with useful, experimental, and fun services. Currently, they operate a variety of services that have been launched to students at the university, and have several new projects in the works.

Blockchain Valley 

Blockchain Valley is a blockchain society based at Korea University. Launched in March 2022, it currently has 54 members active in development, research, security, and validator teams. Centered around product development, research paper publication, zero-day research, core research, and validator operations, the society maintains active engagement through collaborative projects with other societies, corporate partnerships, and participation in ideathons and hackathons.

X: @blockchainkor

BlockBlock

BlockBlock is a blockchain club at Yonsei University founded in September 2022 with the hope of bringing together people who want to study blockchain together. We strive to become an ecosystem where people interested in blockchain can gather and grow to positively impact the WEB3 industry. True to the spirit of blockchain, BlockBlock is run by people from various departments who bring diverse perspectives.

X: @ysblockblock

CURG

CURG is a blockchain consortium established in 2018, primarily among universities in the Seoul metropolitan area. Composed of tech enthusiasts, CURG conducts blockchain research and Proof of Concept (PoC) projects with a 3T mindset: Trendy, Thoughtful, and Transdisciplinary. Through this, it aims to accelerate the future of blockchain and contribute to the development of its ecosystem.

X: @curg_official

The architects of the original web specification decades ago were forward-thinking, reserving the 402 status code in anticipation of a future need for native payment standards. That future has come, rolled out by Coinbase.

X402 has arrived: the on-chain payment gateway for AI and APIs. This revolutionary protocol finally activates the sleeping giant, turning that long-unused code into a powerful engine for instantaneous, blockchain-based micropayments across the web.

It’s so great that FLock.io couldn’t resist adopting it. We are creating a unified gateway for on-chain Agents to pay for the brains (AI Models) that drive them with X402.

What exactly is X402?

X402 is a protocol that activates the HTTP 402 status code to enable instant, blockchain-based payments for web resources and APIs. It’s presented as the best way to accept digital payments.

It’s an open, internet-native payment protocol developed by Coinbase. Built around the HTTP 402 status code, x402 enables users to pay for resources via API without registration, emails, OAuth, or complex signatures.

A client, such as an AI Agent, requests a paid resource from a server, such as an API. the server responds with an HTTP 402 status code. The 402 response includes the payment requirements (amount, currency, address – often a stablecoin like USDC on a layer-2 chain like Base).

What’s X402 good at?

It offers new monetisation models, offering developers and content creators a frictionless way to earn revenue from small transactions without forcing subscriptions or showing ads.

• Monetise APIs & AI models: charge per API call, AI inference, or data access without subscriptions.

• No chargebacks or middlemen: instant USDC payments with finality, no disputes.

• Built-in compliance & security: KYT screening from Coinbase including OFAC and illicit finance.

• Real-time reporting dashboards: track revenue, balances, and transactions effortlessly.

X402 as a protocol has 0 fees for either the customer or the merchant. It enables instant settlement, accepting payments at the speed of the blockchain. It is blockchain agnostic, meaning it isn’t tied to any specific blockchain or token but is open to integration by all.

It encourages broad community participation and isn’t tied to any centralised provider. Anyone can implement or extend x402.

FLock.io is using X402 for on-chain agents

FLock.io is building the infrastructure for decentralised, privacy-first AI. By integrating X402, we are solving the payment problem for on-chain, autonomous AI agents. With FLock API and X402, there’s going to be an explosion of applications powered by FLOCK, our token. 

For example, the AI astrologist, vibe coded using the FLock stack, can be found on vibe.flock.io. It’s also a Base app. Check it out while registration is open!

X402 is the perfect fit for FLock.io and on-chain AI agents. We are creating a unified gateway for on-chain Agents to pay for the brains (AI Models).

AI Agents are designed to operate independently, executing complex, multi-step tasks across the web. They need to be able to:

1. Read an article (pay a micro-toll).

2. Access an API (pay per-inference).

3. Use a decentralised computing service (pay per-second).

4. Buy a piece of data (pay per-query).

X402 is a privacy-first way of paying. This aligns perfectly with FLock’s core commitment to decentralised AI development. Read our recent blog to find out why blockchain isn’t a cliché at FLock.io – it’s essential for scaling federated learning.

No personally identifiable information (PII) required by X402. Agents and users transact using only their wallet address. There is no registration, no email, no OAuth, and no collection of PII required to make a payment.

The x402 protocol is stateless and works purely on HTTP headers and blockchain transactions. This is the only way to facilitate the high-frequency, autonomous payments that a fully operational AI agent requires.

More about FLock.io

FLock.io’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer and AI model marketplace. We also just launched our API Platform.

Find out more about FLock.io by reading our docs and blogs. See a sneak peak of new launches coming in Q4 2025.

For future updates, follow FLock.io on X.

See our docs for our updated onboarding guide for AI Arena. Get ready to stake, train and earn on train.flock.io!

What has changed in V2.1?

AI Arena V2.1 has introduced two major changes to the reward structure.

Change 1: Fairer rewards for training nodes and validators that receive a delegation

Before the V2.1 upgrade there was a problem: training nodes and validators saw their personal rewards reduced when they accepted delegations because they had to share the same percentage pool.

V2.1 fixed this inherent unfairness to properly reward those training the model. We now use two separate reward percentage arrays: one with delegation and one without.

1. With delegation: Total reward percentages involving delegators.

2. Without delegation: Reward percentages solely for training nodes/validators.

The difference (1 - 2) represents the share between training nodes/validators and delegators, adjusted by the reward sharing ratio.

This means training nodes and validators no longer receive fewer rewards for having delegation. Consequently, it also means that delegators may have fewer rewards because now they are only eligible for the rewards attributed to the delegated amount.

Hovering over Claim FLOCK shows the Daily Rewards, Unlocked Vested Rewards and Pending rewards that are still under vesting

Change 2: Less ‘sell pressure’ on final day of task

Training nodes can also expect less ‘sell pressure’ on the final day of each task. We achieve this by only letting training nodes claim final rewards gradually.

Previously, 95% of rewards were unlocked in one go on the final day, and 5% were released on a daily basis. 

After this upgrade, 95% of rewards are instead unlocked gradually after the task has ended, with  the rewards being vested over 30 days. 

Vested rewards are adjusted every second, promoting stability and rewarding long-term commitment. There is no dilution regarding vested rewards. 

What do users need to do?

Nothing! With the update, changes are only made to the way rewards are distributed. How you stake, delegate and interact with pools remains intact.

What is AI Arena?

AI Arena is a platform for competitive model training. Decentralised AI leader FLock.io launched AI Arena on train.flock.io in May 2024. It brings Web3 a leap closer to traceable contribution and on-chain incentivisation for data owners, model developers and compute providers.

See you in the arena!

We can’t wait to see the V2.1 upgrade promote long term and sustainable growth and development within the FLock ecosystem. The team will be working closely with our community through Discord, Twitter, and developer channels. Head over to train.flock.io. 

A huge thank you to all the FLockies for being a part of this project!

More about FLock.io

FLock.io’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer and AI model marketplace. We also just launched our API Platform.

Find out more about FLock.io by reading our docs and blogs. See a sneak peak of new launches coming in Q4 2025.

Read our recent blog to find out why blockchain isn’t a cliché at FLock.io, it’s essential for scaling FL.

For future updates, follow FLock.io on X.

We are co-leading seven global initiatives in Cohort 2 that are showcasing FLock.io technology to deliver instant financial relief, ensure community benefit-sharing, and build transparent governance in some of the world’s most vulnerable regions.

They are building powerful use cases of blockchain-powered decentralised AI (deAI) and federated learning (FL) technology to drive meaningful impact across the developing world from climate action and finance to healthcare, wildlife conservation and cultural preservation.

As the United Nations Development Programme’s (UNDP) AI Strategic Partner, we cannot wait to see where these initiatives go. They strive towards the UN’s Sustainable Development Goals (SDG) by tackling challenges in Liberia, Rwanda, Latin American & Caribbean region, India, Morocco, and Sierra Leone.

Here are the details of the seven cohort 2 projects we are representing.

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

Harnessing Africa’s natural wealth: wildlife economy in Rwanda

NFT-based conservation financing platform combining digital collectibles, VR experiences, and gamification to replace tourism-dependent wildlife funding and empower local communities.

This project tackles the critical funding crisis facing gorilla conservation in Rwanda, which was severely impacted by a 70% drop in tourists due to COVID-19. Threatening both vital habitats and community livelihoods, this challenge required an innovative, sustainable solution. 

In collaboration with blockchain researchers from the University of Cambridge, this project develops an NFT-based conservation financing platform designed to replace tourism-dependent funding. By combining digital collectibles, VR experiences, and gamified engagement, the platform generates ongoing conservation revenue, ensures equitable benefit-sharing with local communities, and follows a Pan-African roadmap for scalable replication. 

This solution directly aligns with:

• SDG 4 (Quality Education)

• SDG 8 (Decent Work and Economic Growth)

• SDG 15 (Life on Land).

Blockchain for inclusive climate and gender finance in Latin American & Caribbean region 

Blockchain-enabled parametric micro insurance and gender responsive finance platform protecting smallholder farmers and women-led Micro, Small, and Medium-sized Enterprises (MSMEs) from climate shocks and financial exclusion.

Addressing persistent financial exclusion and high climate vulnerability across the Latin American and Caribbean region (specifically focusing on the Dominican Republic and Peru), this team, also partnering with the University of Cambridge, is developing a unified blockchain-based financial platform. 

The solution integrates parametric climate insurance, gender-responsive lending, and micro-savings into one accessible system, specifically targeting women, youth, and informal workers. 

Built on transparency and digital identity, the platform is engineered to deliver insurance payouts within a critical 72 hours after climate events, support AI-enhanced financing for women entrepreneurs, and build long-term financial resilience for the unbanked and underbanked. 

The pilot targets 500-1,000 farmers ((≥ 4 0% women, ≥ 20% youth). It aims to reduce income loss by 20%, and it deploys ≥ 30% idle reserves to lower premiums.

This initiative strongly supports: 

• SDG 5 (Gender Equality)

• SDG 8 (Decent Work and Economic Growth)

• SDG 13 (Climate Action).

Blockchain for patient-centred governance in Mauritius & Seychelles

A blockchain-enabled system that empowers patients in Mauritius with transparent control over health data, aligning with Sustainable Development Goals (SDG).

This project focuses on the foundational challenge of patient trust in digital healthcare, particularly in Mauritius and Seychelles, where lapses in consent, transparency, and compliance risk undermining the system. Working with researchers from Lingnan University Hong Kong, the project is building patient-controlled data sharing and audit mechanisms. 

The solution leverages blockchain and federated learning to enable real-time consent management, significantly reduce legal risks for healthcare providers, and ensure compliance with the Mauritius Data Protection Act 2017. 

It allows patient notifications, completely prevents unauthorised access and allows 24/7 audit transparency.

By setting new standards for privacy and digital healthcare, this mechanism advances:

• SDG 3 (Good Health and Well-Being)

• SDG 10 (Reduced Inequalities)

• SDG 16 (Peace, Justice and Strong Institutions).

Empowering India’s rice farmers through blockchain carbon credits

For millions of smallholder rice farmers in India, access to global carbon markets remains out of reach due to expensive, time-consuming, and intermediary-controlled verification processes. This exclusion prevents farmers from benefiting directly from the climate value they create through sustainable practices. 

To solve this, FLock.io supported the development of a blockchain-based MRV (Measurement, Reporting, and Verification) system with NovaChat for UNDP India. This platform automates the tracking of field practices, quantifies emission reductions, and streamlines proof submission. By automating the MRV process and ensuring transparent data flow, the system enables farmers to directly access carbon credit income, effectively turning climate-smart agriculture into a sustainable livelihood opportunity. 

It engages 3,000 farmers across 2,500 hectares. The methodology is applicable to 44 million hectares of rice paddy in India. It targets 30% emission reduction.

This initiative is key to advancing: 

• SDG 1 (No Poverty)

• SDG 10 (Reduced Inequalities)

• SDG 13 (Climate Action).

AI+Web3 SDG solution for Moroccan crafts, people and artisans

AI and Web3 platform empowering Moroccan artisans through NFT authentication, digital marketplaces, and global market access to increase income and preserve cultural heritage.

• Multi-modal AI creates marketing content (text/image/video) for artisans lacking digital skills.

• Direct market access eliminates intermediaries, increasing artisan compensation.

• Omnichannel integration connects artisans to global eCommerce, social media, and Web3 platforms.

This project supports Morocco’s artisans, who, despite preserving centuries of heritage, often suffer from low incomes, fragile value chains, and limited access to global markets. Teaming up with MesuAI, this initiative is building an on-chain platform that empowers artisans to authenticate their work, share their unique stories, and sell directly to a global audience.

There is an $8.3 billion global TAM (total addressable market), $680 million global SOM (serviceable obtainable market) and $8.3 billion domestic SOM.

This initiative aligns strongly with: 

• SDG 1 (No Poverty)

• SDG 8 (Decent Work and Economic Growth)

• SDG 9 (Industry, Innovation and Infrastructure).

Sierra Leone carbon ledger initiative

Addressing critical gaps in data, access, and oversight that have marginalised communities from benefiting from global climate finance in Sierra Leone, this project partners with NovaChat to build a custom national carbon registry. 

Developed during the accelerator, the system is built on Avalanche and is designed to enable national emissions tracking, implement robust MRV (Measurement, Reporting, and Verification) systems, and ensure equitable benefit-sharing across Voluntary Carbon Market projects. 

By strengthening transparency and ensuring fair distribution of climate finance, this initiative aligns with the Paris Agreement and positions Sierra Leone as a regional leader in climate action. 

• Smart contract automation enables instant DSA disbursement in USD-pegged stablecoins.

• Immutable blockchain records prevent ghost beneficiaries and enable real-time fraud detection.

• Triple-access system supports mobile apps, USSD for basic phones, and paper vouchers for non-phone users.

• One-time KYC with face-photo authentication streamlines participant verification.

• Offline-first architecture designed for nationwide and West Africa scalability.

It distributes 20% of verified project proceeds to communities and there is $6-15m addressable historical profit through private VCM projects.

This solution is crucial for advancing: 

• SDG 10 (Reduced Inequalities)

• SDG 13 (Climate Action)

• SDG 17 (Partnerships for the Goals).

Solving Liberia’s DSA payment challenge with blockchain wallet

The project establishes a blockchain-based payment architecture—supporting smartphones, feature phones, and paper vouchers—to instantly and transparently deliver Daily Subsistence Allowance (DSA) to workshop and training participants in Liberia, radically reducing payouts from up to a week to near-real-time.

It addresses a persistent issue for UN event participants in Liberia: the slow, manual process of receiving Daily Subsistence Allowances (DSA), which often takes days and relies on paper-based verification. 

In collaboration with Korea University, this team is co-designing a blockchain-based system for instant, transparent DSA disbursement across the country. 

The solution uses smart contracts to automatically release USD-pegged stablecoin payments. Participants can check in via QR code, biometrics, or USSD, including crucial offline options for areas lacking reliable connectivity. 

By cutting payout times from days to near-real-time and utilising immutable blockchain records to prevent fraud and ensure accountability, this project establishes a foundation for scalable, inclusive digital finance in West Africa. 

• Smart contract automation enables instant DSA (Daily Subsistence Allowance) disbursement in USD-pegged stablecoins.

• Immutable blockchain records prevent ghost beneficiaries and enable real-time fraud detection.

• Triple-access system supports mobile apps, USSD for basic phones, and paper vouchers for non-phone users.

• One-time KYC (Know Your Customer) with face-photo authentication streamlines participant verification.

• Offline-first architecture designed for nationwide and West Africa scalability.

It offers a near-instant payout, reducing the delay from up to 1 week. There is a 12-week development roadmap for MVP pilot. 

This initiative directly supports: 

• SDG 8 (Decent Work and Economic Growth)

• SDG 16 (Peace, Justice and Strong Institutions)

• SDG 17 (Partnerships for the Goals).

About the SDG Blockchain Accelerator at the UNDP

The SDG Blockchain Accelerator is a global initiative that helps scale blockchain and AI solutions to advance the SDGs. Strategically led by UNDP’s Alternative Finance Lab (AltFinLab), in partnership with the Blockchain for Good Alliance (BGA), EMURGO Labs, FLock.io, Stellar Development Foundation and Sui Foundation. 

The Accelerator connects innovators with real-world challenges to turn ideas into impact. It is made possible through catalytic funding from the Cardano treasury, through Project Catalyst, alongside contributions from other partners.

More about FLock.io

FLock.io’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer and AI model marketplace. We also just launched our API Platform.

Find out more about FLock.io by reading our docs and blogs. See a sneak peak of new launches coming in Q4 2025.

Read our recent blog to find out why blockchain isn’t a cliché at FLock.io, it’s essential for scaling FL.

For future updates, follow FLock.io on X.

It’s becoming increasingly evident that simply anonymising and de-identifying sensitive data is no panacea for big tech’s data privacy problem. In the case of a data breach, someone can easily use patterns to re-identify the data, especially with such enormous and rich datasets.

Concerns are continuously mounting for patients’ data privacy. In previous years, AI was primarily used for administrative streamlining. Now, the top application is processing or analysing sensitive medical data, which is deeply troubling given how insecure today’s centralised AI systems are.

And yet, it’s easy to see the allure of centralised AI. Healthcare systems worldwide have buckled. The pandemic exposed deep vulnerabilities in medicine, and hospitals are sprinting to catch up.

AI offers an enticing fix: early disease detection and drug discovery to save more lives. The public sector aims to spend less time on administration and more time delivering the services people rely on.

The question is: should patient data be shared with big tech to create an AI-driven NHS?

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

The power of AI in action

The results of AI are already tangible. In the NHS, hospitals using an AI agent for admin have treated 114 more operating theatre in-patients every month. Last year, AI saved a stroke victim’s life by pinpointing the exact location of a blood clot in her brain in a medical emergency, sparking calls for it to be rolled out nationwide.

Prime Minister Keir Starmer praised it as “the power of AI in action” in a speech about the government’s plan to train models on patient data. He hopes it will “predict and prevent” strokes in the future.

In January, the government set out a blueprint to turbocharge AI and unleash it across the UK. The NHS is in the midst of its 10-Year Plan to go from paper to digital. It’s building a world-first AI system based on its IT software, the NHS Federated Data Platform. AI is currently being piloted to help free up hospital beds being used by people who are fit to be sent home.

The brand new AI stethoscope is another example of a brilliant piece of technology that can detect heart conditions in seconds. Developed by Imperial College London, it enhances a classic medical tool that was invented in France in 1816 and upgrades it to the 21st century.

But the dangers of power concentration in medicine have been well documented within big pharma. During WW2, pharmaceutical firms promoted and sold methamphetamine to soldiers. When patients’ data is concentrated, we may face similar manipulation driven by profit.

The vulnerabilities of health data

Centralised data servers are a double-edged sword in AI, particularly in medicine. The core issue is not what it does, but how it works. The medical data is sent to a massive, central cloud server to be analysed by AI trained on data from tens of thousands of patients. These servers are often concentrated to large tech conglomerates that possess and train the models, highlighting significant security risks.

Sending sensitive patient data from private hospital servers to the cloud for analysis should sound alarm bells. The patient and the healthcare system lose data control as soon as it leaves the hospital system. This single point of failure significantly increases the real possibility of disastrous data breaches.

The vulnerabilities of health data have already been exemplified in recent ransomware attacks that sent the NHS reeling. A cyberattack against a pathology laboratory, Synnovis, caused colossal disruption to NHS hospitals in London last June. The Russian cyber-criminal group Qilin shared almost 400GB of private information on their darknet site. The attack led to the postponement of over 1,000 operations and 3,000 appointments and disrupted blood services, with a patient death later linked to the incident.

Ultimately, the goal should be to achieve the life-saving benefits of AI, but without the data privacy trade-off. The NHS now has comprehensive electronic patient records, making it well-positioned to perform impactful health research and prevent public health risks.

Hospitals should be empowered to leverage algorithms that can identify cancers in medical imaging with accuracy that exceeds human radiologists. They should harness AI, trained on diverse datasets, to reduce healthcare inequalities – gender bias can be catastrophic when identifying heart attacks in women.

Looking to decentralised AI

All this and beyond can be achieved through Federated Learning (FL), a form of decentralised AI (DeAI). It allows collaborative model training without exposing raw data. Instead of data moving to the model, the model (or a version of it) moves to the data.

FL is a privacy-preserving, GDPR-compliant alternative to the big tech “black box” of traditional centralised AI dominating the market. The core benefit is that sensitive patient data never leaves its secure, local environment (a hospital’s server). The NHS would maintain full control over its data, and wouldn’t need to relinquish its most valuable asset to a third-party corporation.

By training models on diverse datasets from many different hospitals, which serve diverse populations, the final model is less prone to biases. This leads to more equitable healthcare outcomes.

While the NHS puts the Foresight project on hold, governments like the UK should listen to calls for a decentralised future that prioritises innovation that isn’t at the expense of anyone’s personal data.

About the author

Jiahao Sun is the founder and CEO of FLock.io, a decentralised AI platform. An Oxford alumnus with expertise in both AI and blockchain, he previously served as director of AI for the Royal Bank of Canada and an AI Research Fellow at Imperial College London. Sun founded FLock.io to pioneer privacy-centric AI solutions, aiming to democratise AI development through secure, collaborative model training and deployment.

September saw the historic UK-US tech prosperity deal signed by Donald Trump and Keir Starmer. The pact “broke all records,” the Prime Minister said.

On the face of it, it’s a watershed moment. Scores of investments and partnerships are to be injected into Britain. It’s focused on building new data centres and growing AI start-ups and cutting-edge tech, and developing advanced quantum computers. 

Nvidia CEO Jensen Huang said back in June that while Britain is having a goldilocks moment for AI, it lacks the computing power it needs to capitalise on its potential. He added: “It is surprising this is the largest AI ecosystem in the world without its own infrastructure.”

The tides have seemingly turned for Britannia. Nvidia, OpenAI and Microsoft are pouring an extraordinary amount of cash into AI infrastructure. It will turn the North East of England into a new AI Growth Zone with 5,000 new jobs. Several enormous data centres are planned with a helping hand from BlackRock, as well as the country’s most advanced AI supercomputer. 

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

Starmer has made AI in healthcare a priority area this year, and he hopes the deal will catalyse medicine breakthroughs at a time when the NHS is crumbling. “What a day. 250 billion pounds flowing both ways across the Atlantic,” he said. “It is the biggest investment package of its kind in British history by a country mile.” His government pitched it earlier this year as vital in the tech race against China.

But all is not as it seems – the UK could be giving up AI sovereignty for American compute. The pledges from American firms are far from charity donations, and carry their own subtext. The digital infrastructure will predominantly serve US tech giants, not necessarily British projects, and cement the dominance of “black box” centralised AI.

AI requires massive amounts of electricity, and Trump has declared a national energy emergency to approve coal power plants to fuel it. But it’s not enough – US big tech needs data centres on foreign soil too. Britain is set to become populated with energy-hungry data centres that run American models, with dire environmental consequences to follow. One new Google data centre in Essex is expected to emit the equivalent of about 500 flights from Heathrow to Málaga every week. The idea that governments can become AI leaders and simultaneously meet net zero targets amounts to “magical thinking at the highest levels,” as one University of Cambridge report puts it.

With companies like OpenAI, the pressure to innovate competitively and deploy quickly clashes with the need for caution. Centralised AI is plagued by its black box problem of opaque decision-making and faltering public trust. All data is sent to a central server, creating a single point of failure and posing hazards for data privacy, security and user autonomy.

At a time when the UK has postponed plans to introduce AI regulation until at least the summer of 2026, the plans risk derailing the country’s trajectory to become a maverick AI leader that refuses to compromise ethics for the sake of innovation. Britain’s own diverse ecosystem of independent projects, many of which prioritise safety over profit, are endangered by the looming Silicon Valley monopoly.

The US encourages open innovation with free market capitalism whereas Europe has a risk-based approach. Post-Brexit, the deal may steer the UK in the direction of the Atlantic Ocean. Washington is currently applying pressure on the UK to change two pieces of regulation. Firstly, the Digital Services Tax, which Trump claims discriminates against US companies. Secondly, the Online Safety Act, which he portrays as anti-democratic censorship.

In his book Vassal State: How America Runs Britain, Angus Hanton lays bare the extent to which US corporations own and control Britain’s economy. It argues that in the supposed “special relationship,” the UK has ironically become America’s colony rather than its economic partner. To Hanton, policymakers shape their decisions to suit the whims of their American corporate overlords.

It’s precisely this sentiment that Nick Clegg, the former Liberal Democrats leader who spent six years at Meta, echoed in reaction to the deal. He decried the pact as “mutton dressed as lamb” and “leftovers” from Silicon Valley. The UK is becoming ever more reliant on “Uncle Sam’s coat-tails” instead of scaling its own independent tech sector. It’s not flowing both ways, but rather is “all one-way traffic,” he added.

One project is welcome to an extent: The Stargate Project. This will give the UK local computing power to run OpenAI’s models in sensitive areas such as the public sector, financial services, national security and scientific research. In this way, data will not need to be sent abroad. But while the promise of the Stargate Project to run models locally is a step toward data security, but it’s not data sovereignty. Decision-making still takes place at Sam Altman’s desk behind closed doors, and sensitive data is still sent to a central server at a data centre.

The notion of AI sovereignty “borders on meaningless” because so many businesses already “rely on American cloud companies”, wrote Bloomberg columnist Parmy Olson. “The US owns about 75% of the world’s AI supercomputers.”

The narrative pushed by the UK-US tech prosperity deal – that massive, centralised infrastructure is the only path to AI leadership – is fundamentally flawed. We no longer have to choose between powerful models and ceding control to Silicon Valley giants. With the emergence of decentralised AI (DeAI) and techniques such as Federated Learning, there is a high-performance, sustainable alternative that respects data privacy, the community and the planet. 

DeAI operates closer to the data source, cutting the energy costs. Local computation happens on devices like smartphones and local servers, diminishing by much more the amount of data that needs to be sent over to a centralised location, and further cutting power consumption. This ensures that datasets remain secured within their owner's firewalls, delivering true data sovereignty while models are collectively enhanced.

DeAI is part of the broader mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent, and composable development. This is where projects like FLock.io come in.

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer. We just launched our API Platform.

Find out more about FLock.io by reading our docs. See a sneak peak of new launches coming in Q3/Q4 2025.

For future updates, follow FLock.io on X.

It’s an easy way to create API keys to use with AI models. Just sign up, generate an API key, and you’re ready to start making calls to the models available on FLock.

It’s OpenAI SDK-compatible, so if you’ve used OpenAI’s API before, you can get started immediately and intuitively with minimal learning curve. It makes it easy for developers to seamlessly build, integrate and scale AI-powered applications.

This new launch will be integrated with Moonbase, our rewards layer, via which the revenue will be distributed. It’s the latest example of how the FLock ecosystem is becoming more interconnected and incentive-driven.

Stay tuned to witness this launch propelling collaboration, innovation and real-world impact in DeAI. It’s a fantastic year for FLock so far, and more is on its way – check out this sneak peek of Q3/Q4 launches.

FLock’s API Platform features

The built-in analytics tracks the usage and logs of each model, giving smart insights into demand and performance. By creating a feedback loop of train → deploy → use → reward, we’re guiding developers towards the models the community needs most.

Our API Playground feature is an interactive sandbox for safely trying out API calls to see how they work, without writing any code. You select an API key and a model, then start a conversation.

Our AI Platform isn’t just for solo developers – it’s also designed for teams. From the dashboard you can add new members for collaboration and sharing, and allocate a budget. This makes it easier for startups, research groups and organisations to scale their projects together. 

See our docs on docs.flock.io for a step-by-step guide.

Roadmap: More API Platform features coming soon

FLock has exciting plans to further develop the API Platform after this initial release.

Further down the line, the highest performing AI Arena models will be deployed to the API Platform for real-world use. This means the fiercest models will have even more potency to make an impact.

Several models are already available, with fine-tuned models from our flagship DeAI training platform AI Arena, found on train.flock.io, landing in the coming weeks.

Specifically, in near future, Moonbase will serve as another essential role: a distributor of revenue to those who contributed to AI Arena-trained models hosted on the API Platform. A revenue loop will flow earnings back into Moonbase, funnelling rewards to developers and creating a direct incentive loop.

All these additions will further incentivise training on train.flock.io, distribute equitable rewards and make the FLock ecosystem more interconnected and self-sustaining.

Currently, users can start conversations on the API Platform, but in the future users will also be able to generate images.

Our API Platform interface is easy to use no matter your experience level: request an API key and make your first call. Head over to platform.flock.io.

Step-by-step guide

Read our docs for the full instructions.

Step 0: Head to platform.flock.io and sign up

Step 1: Create an API key

To start using the API, you will be prompted to create the API key first. It uses API keys for authentication. Keep it safe — you won’t be able to see it again.

Then select all the models you wish to access via the API and click on "Create API"

Please keep your API key safe as you won't be able to see it again. If you lose it, please create a new key.

Step 3: Use the API

Please refer the section API Endpoints

Step 4: Buy credits

You can buy credits under Setting > Billing

We support payments using Stripe and Base

Step 5: Track Usage

Under the Usage tab, you can track the total requests, total tokens and average cost per request

Step 6: Adding members to your organisation

Under Teams > Members , you can add new members to your organisation

Step 7: API Playground

You can try out different models directly in the API Playground

Read our docs here and access AI Arena on train.flock.io 

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

We are part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on X.

A wave of new data centres is being planned, and their staggeringly huge water and electricity needs risk derailing net zero. One new Google data centre in Essex is expected to emit the equivalent of about 500 flights from Heathrow to Málaga every week, and it’s one of many being built.

Multibillion-dollar data centre deals with tech giants like OpenAI and Nvidia are expected to be announced during Donald Trump’s state visit to London this week. The US and UK are hurrying to finalise a trade agreement as they race to become AI trailblazers, but dire environmental consequences will follow. 

When you save your photos and videos in the cloud, it sounds weightless, but that couldn’t be further from the truth. Your files are stored in huge buildings full of GPUs that in 2023 alone reported an electricity consumption worldwide comparable to some mid-sized countries.

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

Data centres have been around for decades, but the current boom is driven by AI. Modern ML models are far more computing power-hungry than the internet. A single ChatGPT query needs 10 times as much electricity as a Google search.

To call data centres thirsty is an understatement. Without their water cooling systems, they would instantly overheat. To put it into perspective, generating just 100 words with GPT-4 uses three bottles of water. Large data centres can consume up to 5 million gallons per day, equivalent to the water use of a town populated by 10,000 to 50,000 people.

The idea that governments can become AI leaders and simultaneously meet net zero targets amounts to “magical thinking at the highest levels,” as one report from the University of Cambridge puts it. Even the most conservative estimate for big tech’s energy needs will see a five-fold increase over the next 15 years.

While Taylor Swift gets bashed for her jet set carbon footprint, it’s actually data centres that will far exceed global air travel when it comes to greenhouse gas emissions. They currently account for 1.5 per cent but are projected to reach 8 per cent by 2040. In some countries, the electricity use is shocking – an enormous 20 per cent of all power in Ireland now goes to Dublin’s cluster of data centres.

There have been some efforts to reduce the colossal power needs. Google has experimented with its DeepMind AI to improve data centre efficiency and mitigate the environmental impact. However, with the whole army of centres being planned, mere mitigation won’t cut it.

There is a greener alternative. With traditional, centralised AI like ChatGPT, data is sent to a central server at a data centre for processing. Decentralised AI (DeA) eliminates the central server altogether. 

Instead, it taps into a distributed network of often idle GPUs across various geographies. This approach, leveraging techniques like federated learning, could dramatically minimise centralised server power consumption. 

DeAI operates closer to the data source, cutting the energy costs. Local computation happens on devices like smartphones and local servers, diminishing by much more the amount of data that needs to be sent over to a centralised location, and further cutting power consumption.

It will be grossly irresponsible for BlackRock to be allowed to go ahead unbridled with a £500m investment in UK data centre infrastructure, and for OpenAI to launch the UK arm of its £400bn Stargate centre.

Let’s instead follow the path towards a more sustainable AI ecosystem. By moving AI training away from centralised, energy-intensive data centers and toward a decentralised, distributed model, we can continue to innovate while also protecting our planet.

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

FLock is part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on X.

Likewise, the gold rush for Web3 crypto rewards is driving the initial uptake of decentralised AI. It’s key in the battle to win public opinion and fight for sovereignty over our data as “black box” centralised AI dominates the market.

Big tech monopolies have tons of resources at their disposal – Sam Altman says OpenAI is sitting on more than 1 million GPUs. But centralised AI carries a risk of data abuse, and users have no say or ownership.

Privacy-preserving AI is possible, and it’s landed. However, unlike big tech, with DeAI there’s no single corporation with deep pockets and a huge data centre. But there’s no need – because tapping into a global network of expertise, idle computing power and diverse data is actually vastly more convenient, efficient and innovative. 

This blog explores how on-chain crypto tokens fuel DeAI ecosystems and how blockchain makes them so effective.

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

What exactly are crypto tokens in DeAI?

Crypto tokens in DeAI were born out of the match made in heaven between AI and blockchain. Crypto incentives in DeAI are not the same as cryptocurrencies, although both rely on blockchain. Cryptocurrencies are like digital money. Even Starbucks used to accept it as payment. They are native to their own independent blockchain, such as ETH on Ethereum.

Usually, crypto tokens are built on an existing blockchain – for example, FLock’s native token $FLOCK is on Base, an Ethereum Layer-2 blockchain. It's a utility token designed specifically to be used within an ecosystem or dApp – it cannot buy frappuccinos in the real world.

Crucially, DeAI crypto incentives do have liquidity on a global market. For example, a contributor who earns $FLOCK can then go and sell it on cryptocurrency exchanges. Other tokens include TAO (Bittensor), OCEAN (Ocean Protocol) and AGIX (SingularityNET).

Crypto incentives ensure a steady stream of high-quality contributions and ward off bad behaviour – and blockchain provides the underlying foundation upon which these on-chain tokens live.

How blockchain makes crypto incentives so effective

With traditional AI, the pressure to innovate competitively and deploy quickly clashes with the need for caution. It is plagued by its “black box” problem of opaque decision-making, privacy issues and faltering public trust.

There’s an alternative to trusting an opaque system to manage rewards. The reason blockchain is called “trustless” is because it doesn't require trust – it’s secured by code instead. 

Usually associated with cryptocurrencies like Bitcoin, AI has given blockchain a new lease of life. It promises the security, immutability and transparency that AI lacks, and creates a automated system that distributes the incentives fairly.

Blockchain transparently records everything on-chain and cannot be altered. Contributors can be confident that the tokens they earn are permanently recorded, and cannot be taken away. Smart contracts automatically distribute tokens once their work has been verified by the network. This is fairer and less prone to corruption because it’s managed by code, not a human.

Incentives drive sustainable contributions of data, compute & more

The collective, decentralised pool of resources and expertise makes DeAI a force to be reckoned with. On-chain crypto rewards are the lifeblood, incentivising a stream of high-quality contributions. These includes: data, computing power, model and algorithm innovation, governance, inference and validation.

Tapping into a global network of idle computing power is vastly more efficient than building a massive data centre. Individuals with spare GPU or CPU power on their computers can rent it out to the network. Their devices become nodes that perform the heavy lifting of AI tasks. For example, Akash lets users buy and sell computing resources – its token AKT is used to govern, secure the blockchain and incentivise participants.

Instead of a single team of researchers and engineers, anyone can contribute to the development of AI models. When a model is used by others in these ecosystems, its creators can earn a portion of the revenue. For example, on FLock’s AI Arena, users can participate as a a trainer; a validator to verify the contributions of others; or they can stake tokens on behalf of other participants and earn rewards in return. 

AI models are only as good as the data they are trained on. In big tech, a company like Meta collects vast amounts of user data. In DeAI, individuals or organisations can contribute their own data for rewards in a privacy-preserving way, and it’s often more diverse and less biased.

Many tokens also give users a part in on-chain governance, giving them the right to vote on how the project evolves. This is a world away from traditional AI, where all decisions are made internally by the team of executives.

Incentive mechanisms enforce the rules

In a decentralised system, there’s no central authority to enforce the rules. That’s the big challenge of a blockchain-based system: it requires active participation and cooperation to be secure and sustainable. In a dynamic, open environment, human behaviour is unpredictable. 

Incentive mechanisms are essential for regulating that behaviour. A token economy ensures that it’s fair, secure and productive, with the ecosystem operating autonomously.

Blockchain consensus mechanisms include proof-of-work (PoW) and proof-of-stake (PoS). In PoS, participants, called validators, lock up (or “stake”) their crypto assets as collateral to validate new transactions and create new blocks. In PoW, miners competitively solve the puzzles to create a new block and are rewarded with new cryptocurrency and transaction fees, encourages participants to contribute their computational power to the network’s integrity and security.

Crypto tokens can also deter bad behaviour through mechanisms such as “slashing”. If a participant is identified as acting against the system’s rules, their tokens are taken away and rewarded to the honest participants. Slashing protects the system from threats by disincentivising malicious actors and reinforces a culture of trust and cooperation among participants.

FLock wrote an academic paper about peer-to-peer voting and reward-and-slash mechanisms, powered by on-chain smart contracts, to detect malicious behaviours.

More about FLock tokenomics

To make open-source projects more viable, FLock incentivises early adopters with protocol emissions, then shares model revenue with model builders. 

$gmFLOCK, a non-transferable utility token, is the core innovation of FLock’s tokenomics. Participants in the FLock’s AI Arena, such as training nodes and validators, are required to contribute computing or storage resources to complete model training and validation in order to receive rewards.

Read our docs to learn more about how FLock tokenomics powers our ecosystem. For future updates, follow FLock on X.

And yet there is significant appetite across the public sector to make better use of national data to support policy making and improve services through AI, says a Royal Society report. 

Despite the demonstrable benefits of AI in every industry, the privacy risks make it easy to understand why the public sector has been wary of implementing it. Machine learning through Federated Learning (FL) holds the answer.

The problem: traditional AI compromises privacy

Traditional AI sends data to a centralised server for processing, presenting significant security risks due to the single point of failure for all collected data.

Every leak erodes trust – the UK government is currently under fire for a spate of serious data breaches from simple human error, including the bombshell exposure of Afghans who worked with the British military, victims of child sexual abuse and 6,000 disability claimants.

Machine learning (ML) in the public sector is also constrained by data protection laws. UK GDPR’s data minimisation principle requires that the amount of data used is “limited to what is necessary” – but AI systems require extremely large amounts of data to perform well.

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

Federated learning is privacy-preserving machine learning

Federated Learning (FL) is a more GDPR-compliant alternative to traditional ML. It allows collaborative model training without exchanging raw data. FL was developed in 2017 and is an example of a privacy-enhancing technology (PET).

It involves training statistical models over remote devices or siloed data centres, such as mobile phones or hospitals, while keeping raw data localised. It’s a decentralised AI (DeAI) technique. DeAI is a broader concept where processing, data storage, incentives and decision-making are distributed across a network of nodes.

We are on the brink of DeAI’s breakout year, and FL in the public sector is in its proof-of-concept stage – from smart public transport to hospitals collaborating to improve public health.

Figure 1: An example application of FL for the task of next-word prediction on mobile phones (source: https://arxiv.org/pdf/1908.07873)

Let’s look at some public sector use cases that address privacy concerns.

Intergovernmental data sharing

It’s crucial for governments to collaborate to address global problems – and modern solutions often include data-driven methods like AI, which need data. However, data sharing between governments is often hampered by concerns over privacy, sovereignty issues and risks of information misuse.

A paper by Cambridge researchers argues FL as a compelling solution. Governments can create their own AI models and just share the model parameters. This protects privacy while harnessing collective data-driven insights that surpass individual country capabilities.

Humanitarian work is one area where FL could make a huge impact. It could be used on geospatial datasets to predict people’s movements, enabling more effective directing of resources during crises and disasters.

Public health and medical research

Immense life-saving improvements in public health are possible through ML. Algorithms can analyse vast amounts of medical data, including images, patient records, and genetic information, to identify patterns and predict outcomes.

The NHS now has comprehensive electronic patient records, making it well positioned to perform impactful health research and prevent public health risks.

However, doctor-patient confidentiality, strict regulation and data fragmentation across institutions is stopping ML reaching its full potential globally. FL solves this: multiple institutions can collaborate on model development while safeguarding privacy. 

Progress in the right direction is being made. In 2023, the UK and US held a competition for FL use cases to boost pandemic response and forecasting capabilities. Innovators were asked to build solutions for forecasting an individual’s risk of infection.

Oxford researchers developed an easy-to-use technique for hospitals to contribute to the development of AI without patient data leaving the hospital’s premises. This FL pilot used inexpensive pre-programmed micro-computers (Raspberry Pis) in four NHS hospital trusts to train a COVID-19 screening test.

In South Korea, Kakao Healthcare has also delivered a platform that uses FL with Google Cloud to 16 hospitals. In addition, FLock authored an award-winning paper on multi-continental healthcare modelling.

Smart cities and public transport 

Smart city applications such as traffic prediction, water supply management, environmental monitoring and disaster-management require real-time decisions. However, cloud-based computing carries a risk of analytics being delayed due to latency issues, and needs very complex computations.

In Sri Lanka, researchers deployed FL on smartphone accelerometer data to map potholes and road roughness as a developing country use case.

Integrated into the transportation system, smart transportation ensures that traffic movements are convenient, cost-effective, and safe.

Financial crime prevention

Regulators are under pressure to crack down on financial crime, but transaction data is too sensitive to pool. FL allows multiple institutions to spot suspicious patterns without sharing their underlying records.

The UK and US’s 2023 competition also looked at FL use cases that tackle money laundering, which finances organised crime including human trafficking and terrorist financing. Innovators were asked to develop solutions for detecting potentially anomalous payments.

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

FLock is part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on X.

The first half of 2025 saw us named in the 2025 CB Insights AI 100 and launch our Bittensor subnet FLock OFF and the new token gmFLOCK, forge an alliance with Alibaba Cloud’s Qwen, and publish several research papers, to name a few milestones.

Yet more is in store: FL Alliance beta, Moonbase, API Platform, global conferences, a venture capital arm, and task creation support. This blog is a sneak peek into what’s coming.

FL Alliance beta set to get off ground soon

The long-awaited FL Alliance is set to get off the ground towards the end of the year, having opened early registration for the closed beta back in January. 

The beta will start with business customers then expand to retail customers. We are conducting initial phases with DCG and a preliminary phase with AIGEN Sciences in Korea.

Given that “FLock” is actually a portmanteau of Federated Learning (“FL”) and “blockchain”, needless to say we are excited to finally let our love for FL shine through.

FL Alliance will serve as a natural continuation of our flagship DeAI training platform AI Arena. After AI Arena has optimised the model parameters, they are further refined in FL Alliance with proprietary data – read our paper about our robust incentive mechanism.

But what exactly is FL? You can make sense of it as: traditional ML brings the data to the model, whereas FL brings the model to the data. Instead of centralising data on a single server, each participant – be it a mobile device, organisation, or edge server – trains the model locally using their own data. They then share only the model updates (like gradients or weights) with a central server or among each other.

FLock CVC: venture capital arm investing in FLock ecosystem projects

At FLock, we’re passionate about getting quality DeAI projects off the ground. We’re creating FLock CVC – a venture capital arm – to offer that additional funding and support to FLock ecosystem projects.

If you’re a project that uses the platform and allocates rev-shares and token allocations through Moonbase, you have FLock CVC to look forward to. It won’t be an immediate launch but it’s coming.

It’s part of our ethos as a company. We’re keen to cultivate the DeAI and FLock ecosystem by leveraging real-world product-market-fit. By broadening equitable, low-entry barrier participation to independent engineers, we’re turning highly practical grassroots activities into a strongly viable ecosystem of AI agents. 

Moonbase beta launching imminently

Moonbase is a brand-new rewards layer that turns your AI Arena contributions into a launchpad for earning even more. Moonbase lets you stake your FLOCK tokens and unlock a new galaxy of rewards.

This extra layer of incentives lets you earn a variety of tokens, beyond what you’ve already earned on our decentralised training platform AI Arena.

It’s not just for our community – we’re opening up Moonbase to external projects and protocols who want to supercharge their reach and reward new audiences.

In the more distant future, Moonbase will serve as another essential role: a distributor of revenue to those who contributed to AI Arena-trained models hosted on the API Platform. 

AI Arena has already delivered fantastic practical use cases, thanks to the FLock community’s collaboration. So far, 7985 AI models have been trained on the platform. Developers have just finished training an Infini AI concierge for the on-chain card and a CryptoMCP deep research optimised model, to name a few.

Moonbase is one example of how the FLock ecosystem is becoming more interconnected and incentive-driven to propel collaboration, innovation and real-world impact in DeAI.

API Platform as an additional Moonbase feature

The API Platform is almost ready for Q3, to be rolled out as part of Moonbase. It’s an OpenAI SDK-compatible API platform that lets developers seamlessly build, integrate, and scale AI-powered applications.

Head over there soon to create API keys to use with AI models.

FLock has exciting plans to develop the API Platform after its initial release.

Further down the line, the highest performing AI Arena models will be hosted on the API Platform. This means the fiercest models will power real-world applications.

A revenue loop will flow earnings back into Moonbase, which will distribute rewards to those who helped train the AI Arena models. 

It will track the usage and logs of each model, giving clear insights into which models are most in demand, where supply is falling short. This feedback loop between training, deployment and usage will foster a more self-sustaining system.

All these additions will further incentivise training on train.flock.io, distribute equitable rewards and make the FLock ecosystem more interconnected.

AI Agent Marketplace

An AI Agent Marketplace is coming your way at the end of 2025 – a digital platform where users can buy, sell, and deploy DeAI-powered agents designed to perform domain-specific tasks.

It will be distributed along with a privacy-featured Mac Application.

AI Arena V3 task creation support

Currently, only the FLock team can create tasks on AI Arena – we manually create tasks for different model tasks and for different model consumers. We’ve done this deliberately in the beta, keeping things controlled while we refine the system. But with Version 3, we will take things up a notch.

While V2 introduced features like gmFLOCK and delegation pools, V3’s tokenomics is set to enable anyone to create tasks on the platform. Anyone will be able to post jobs or use cases that they envision, and FLock will handle the rest for you. 

It’s a leap forward for decentralisation. We’re progressing AI Arena into a truly open ecosystem driven by users with open participation and distributed control.

It means researchers, developers, businesses, or individuals with an idea can post their own use cases or jobs. Instead of having a gatekeeper, the community itself defines the demand – bolstering resilience and adaptability to real-world use cases, and creating a system less dependent on one team.

FLockCon: Global Conferences for DeAI Builders

FLockCon is our global event series aligned with major blockchain weeks, uniting DeAI builders through theme-based meetups, curated panels, workshops, and hands-on hackathons. Its two goals are clear: 

1. Strengthen ecosystem ties by connecting community members, developers, KOLs, and close collaborators.

2. Attract new builders by inviting more developers and DeAI enthusiasts to join and build with FLock.

Our FLockCon debut will be at Korea Blockchain Week (KBW) Sept 22-28 in Seoul. We’re hosting a builder-focused programme (a “proof of talk” style event in Bittensor terms) featuring panels, fireside chats, and workshops with ecosystem leaders, centered on collaboration and onboarding new contributors to the FLock ecosystem.

We feel lucky to return to Korea so soon. In July, we flew there to host a premier university AI Innovation Challenge with Alibaba Cloud, Qwen and base. 

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

FLock is part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on Twitter.

As a mentor, we will support pilot projects in the Sustainable Development Goals (SDG) Blockchain Accelerator, and its global hackathons – the winners of which will join the next accelerator cohort.

FLock.io will mentor and provide technical support to five pilot projects, helping them to integrate federated learning and decentralised AI. These allow collaboration on sensitive data securely, without moving or exposing it.

Blockchain is leaping ahead in fostering international development, but there’s potential for far more. Our focus will be climate risk modelling, inclusive energy planning, social protection, and transparent supply chains – the next section details the practical use cases the hackathon focuses on.

We are excited to see the life-changing solutions the teams build to anticipate extreme weather; ensure underserved communities access reliable, affordable power; securely analyse financial and welfare data to improve the targeting and delivery of benefits without compromising privacy; and track goods from origin to consumer, reducing fraud and ensuring fair labour practices.

FLock.io is honoured that the UNDP believes our mission aligns with its purposes and principles. This partnership marks a major step in bringing our infrastructure into global real-world use, proving its value, and accelerating adoption.

It’s huge news, and one of several upcoming strategies to utilise the FLock.io tech stack to generate real-world AI-driven use cases, driving real revenue towards our growing ecosystem while making a tangible and benevolent impact. 

FLock.io is a mentor in the global SDG Blockchain Hackathon

FLock.io and other UNDP partners are running the global SDG Blockchain Hackathon from August to September– a highly practical two-week virtual co-creation process to match pressing development challenges. Unlike traditional hackathons, participants will build implementation-ready roadmaps rooted in real-world development contexts.

Teams will explore several blockchain use cases, including:

Financial inclusion: Democratising microinsurance among vulnerable populations like small-scale farmers, and empowering women in Latin America to launch SMEs through alternative credit scoring and lending. 

Conservation & environmental finance: Eliminating double-counting in carbon markets for carbon credit transparency, and creating digital wallets linked to mountain gorillas to enable “interspecies money transfers” in Rwanda.

Healthcare & data privacy: Granting Mauritian patients real-time visibility and control over when their health data is accessed by social services, ensuring compliance with data protection laws.

Supply chain & traceability: Tracking fisheries management systems to provide transparent fish stock data and catch-to-consumer traceability to support sustainable practices.

Financial services innovation: Serving financially excluded remote rural areas through mobile banking agents; enabling smart contracts to automatically trigger payments to farmers upon delivery confirmation in Burundi; and supporting informal workers with transparent micro-lending and skills verification systems in Botswana.

Academia: Blockchain secures academic records in conflict zones; AI enables fair, cumulative assessment beyond final exams.

The hackathon is split into two phases. Phase 1 is a two-week virtual co-creation process matching real development challenges with solution teams. In Phase 2, the top teams join the upcoming accelerator cohort for a four-month build of implementation-ready pilots.

Run by UNDP’s Alternative Finance Lab, the SDG Blockchain Accelerator launched Cohort 1 to back blockchain solutions in climate finance, inclusive energy, social protection, and supply chain transparency. It has a focus on pilots that can be deployed and scaled.

Winners will advance to Cohort 2 in September. Over four months, they’ll refine and scale their solutions with tailored mentorship, technical expertise, and direct access to governments and development organisations.

About the SDG Blockchain Accelerator

The Sustainable Development Goals (SDG) Blockchain Accelerator is a global initiative launched by the UNDP, in partnership with the Blockchain for Good Alliance (BGA) and EMURGO Labs, to leverage blockchain for social impact and advance the UN\s SDGs.

More about the UNDP

The United Nations Development Programme (UNDP) is the United Nations’ lead agency on international development. 

It operates in approximately 170 countries and territories, working to eradicate poverty and inequalities, protect the planet, and build resilient nations. UNDP focuses on helping countries develop strong policies, skills, partnerships, and institutions to achieve sustainable development and the Sustainable Development Goals.

About FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

FLock is part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on Twitter.

The datathon focused on hands-on projects using Qwen models, Alibaba’s open-source LLMs, and FLock’s AI Arena platform on Base. The students worked together on next-generation DeAI applications, leveraging FL to fine-tune Qwen models for specific industry needs.

It was split into two phases – model development then AI application – with 50 and 100 developers joining in respectively, plus professors hopping on board to help out.

FLock partnered with Qwen – a tier 1 open-source LLM – earlier this year, and our milestone Web3 Agent Model is based on Qwen. We were so happy to collaborate again and foster South Korea’s research and talent ecosystem It was amazing to see the awesome next generation of developers have such fun with AI Arena.

Meet our winners!

We chatted with the winners of each track: Sang-min Hwang and Group 6.

1st place on the Base track: Sang-min Hwang

Sang-min Hwang is currently an undergraduate at Korea University. As a member of AIKU, one of the participating partners of the Hackathon, he has been studying and researching AI there.

It seems like you have chosen some specific methods to finish the fine-tuning task. What were those methods and what were your inspirations for choosing them?

I first looked into the dataset, and it seemed like the size of the dataset was smaller than I thought. The given dataset in itself was not big enough for normal fine-tuning, so I focused on how to augment the dataset. I used the Qwen LLM to create synthetic data from the original dataset. Through the augmented dataset, I was able to boost the performance of my fine-tuned model.

What were some difficulties you faced during the hackathon, and how did you overcome them?

The biggest difficulty was dealing with the size of the dataset. As most finetuning methods only work with a big enough sized dataset, I’ve tried both methods that either work on small datasets or larger datasets. For the methods that work on small datasets, I’ve focused more on how to adjust the hyperparameters to enhance the performance. As for the other method, I mostly focused on dataset augmentation to synthetically create a bigger dataset, during which, I used the Qwen LLM, which worked better than I expected.

Q4. What was your most me

morable moment during the whole hackathon?

I’d have to say the most memorable moment for me was when I submitted my model, and a few minutes later, my rank on the leaderboard went from 10th to 1st place. Before I submitted the model, I wasn’t anticipating such a great score, since I thought the changes I made were quite insignificant. The sudden surge in ranks and boost in performance was very exciting and I felt a huge surge of dopamine.

Q5. What were your overall thoughts on the datathon?

I have participated in a couple of datathons in the past, but most of them were held on a centralised platform like Kaggle. It was my first time participating in training in a federated learning way, on a decentralised platform like FLock.io. It was a very novel and cool experience for me, and through the help of the mentors and moderator, it was quite enjoyable, and I feel like it was a good opportunity to level up my skills.

1st place on the Alibaba Cloud track: Team 6

Team 6, a group of three friends from Yonsei University, are members of the university computer club together. Dae-han Lim is an AI major, while Moon-soo Park and Joon Jung are both Computer Science majors.

You three seem to have very good synergies as a team, how did you guys come together as a team, and what’s your secret for such great team work?

I met Moon-soo at YCC, and we took a lot of classes together, where we became close friends. Also, Joon and I were fellow recruits at the military. Since we have studied a lot together at YCC, we developed a very natural team work.

What do you think corporations or the government can do to help produce more AI talents?

If I were to choose one thing, it is ‘money’. 

We live in a very competitive society, and from when we are young, we are taught that the biggest criteria of success is money, especially being in a status with a stable, sufficient income. This is why the top students aim for medical schools, and why a lot of talented students end up being a doctor. They are well paid and it is a very stable profession. 

If AI engineers were paid as well and there was a lot being invested in AI professions and students, talented students will naturally lean more to studying AI.

What were some difficulties you faced during the hackathon, and how did you overcome them?

The lack of time and the number of submissions being limited was our biggest challenge. Although it was quite difficult to think of and try out multiple strategies for enhancing the performance, we had quite a lot of strategies that we have studied together at YCC, which really helped us save time and execute quickly.

You guys won both prizes, 1st place for the Alibaba Cloud track, 3rd place for the Base track, making your team 1st place overall. What do you think was the key to your success?

We approached this datathon through a project management method. We divided tasks and possible methods to each person and meticulously managed the timeline to execute this. 

As this was mostly an online datathon, we thought efficient communication was very important and made sure not to repeat mistakes by always reviewing each trial and error and utilising feedback from past failed attempts as much as possible but also documenting it to always be synced.

What was your most memorable moment during the whole hackathon?

The last day. Even when there was only a few hours left, there were a ton of submissions being made from other teams, which made the leaderboard consistently change until the last moment. We were constantly having meetings to discuss what we can improve more on our final submissions. Although we were short on time, we tried our best to create our best submission and that really paid off when our final submission resulted in us going from 1st to 4th.

If we were to have another datathon, would you do it again?

100%. Especially because this was an online datathon, and not offline, we were free to work on this whenever we had spare time, which really lessened our burden. We also enjoyed that we were able to work on this together as a team.

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

FLock is part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on Twitter.

In a world dominated by massive models and opaque centralised corporations holding all the cards, FLock.io believes there’s room for small, transparent, controllable systems. Models you can fine-tune and run on your own terms.

Both sizes shine in different scenarios – this blog breaks it down.

What’s the difference between small and large language models?

LLMs and SLMs differ primarily in size, computational requirements, and their suitability for different tasks.

As the name suggests, LLMs are massive and often have billions to trillions of parameters. They excel in generative tasks requiring deep language understanding. This means they require significant computational resources for training and deployment, and very few companies have the means to run them. Examples include Gemini, LaMDA, Claude and GPT-4.

SLMs, on the other hand, are much smaller, ranging from millions to a few billion parameters.

The size is often achieved via a technique called distillation (check out our explainer blog), which involves compressing a large model by transferring its knowledge to a smaller, denser, more efficient and lightweight one. It involves transferring knowledge like a teacher to a student. Other methods include pruning and quantisation.

Carry on reading to find out what SLMs are great at.

SLMs shine at speed and efficiency with low resource requirements

SLMs are more lean, compact and efficient than their larger counterparts. As such, they require less memory and computational power, making them ideal for edge computing. Edge computing allows devices in remote locations to process data at the "edge" of the network, either by the device or a local server. On-device AI is made possible, meaning no need for an internet connection or cloud services.

SLMs being lean makes them more efficient, lower cost and less demanding in terms of resources like compute. SLMs include domain-focused models for specific tasks, offering superior accuracy than more broad, versatile LLMs. They are also easier to fine-tune.

SLMs have faster inference and quicker response times because they have fewer parameters to process, making them perfect for real-time applications like virtual assistants.

The tradeoff is that they don’t match frontier performance on generalist benchmarks, have a narrower scope and are more prone to errors in ambiguous scenarios.

SLMs could be the future of agentic AI

The meteoric rise of agentic AI could usher in a mass of applications for which SLMs are sufficiently powerful, inherently more suitable and more economical. 

Most modern AI agents are powered by massive LLMs. Being generalist, they can serve a large volume of diverse requests and are great for broad conversational abilities. However, SLMs are more favourable for repetitive, narrow and simpler tasks. 

Agentic systems can be built with multiple specialised SLMs, and can be built into scalable architecture that is easier to debug and update. Read here about FLock’s new research paper proposing a decentralised agent swarm network. 

Running SLMs locally on devices minimises data transfer, addressing privacy concerns and enabling applications in sensitive areas like healthcare. Read our paper on how blockchain-enabled federated learning has the potential to transform global healthcare.

New research from the Alan Turing Institute backs it up

The Alan Turing Institute (UK’s national institute for data science and AI) has just released new open-source research showing why SLMs are more relevant than ever. The team set out to see how far a small, open-weight language model could be pushed using lightweight tools and without massive infrastructure. 

The results were extremely promising. They experimented on real-world health queries, and achieved incredible performance with a model small enough to run locally on a laptop.

The model had near-frontier reasoning performance. To get there, the team combined retrieval augmented generation (RAG), reasoning trace fine-tuning, and budget forcing at interference time.

Given that medical data is sensitive and difficult to obtain, the team bootstrapped to make it realistic and generated synthetic queries.

With just a few thousand synthetic examples and a bit of test-time reasoning, their lean model, just 3 billion parameters, was suddenly able to engage in thoughtful, grounded reasoning over real-world health content.

Out of the box, small models performed poorly on this task. RAG alone wasn’t enough for interpreting documents or connecting them to user input.

Once they added lightweight test-time scaling techniques, the picture changed. The models became far more capable of drawing logical inferences from retrieved context. In many cases, their responses were nearly indistinguishable from those of much larger frontier models.

The new wave of SLMs

A wave has been released over the past year, typically in the 1B to 8B parameter: Microsoft’s Phi series, NVIDIA’s Nemotron-H family, DeepSeek’s compact distillations, Qwen3’s small models, and Hugging Face’s SmolLLM collection.

Microsoft's Mu is the tech giant's latest lightweight on-device SLM that runs locally on Neural Processing Units (NPUs), aiming to reduce reliance on cloud-based processing.

Get started staking and earning with FLock.io!

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer.

FLock is part of the mission to dismantle the power concentration of AI held by a handful of centralised corporations, promoting more equitable, transparent and composable development.

Find out more about FLock.io by reading our docs. For future updates, follow FLock.io on Twitter.

Governments are hesitant to stifle innovation. Stunned by AI’s transformative potential in their country, world leaders don’t want to miss out on the race. Regulation is lagging behind, with only a handful of nations having passed laws to address the public concerns about AI.

Ahead of Donald Trump’s visit to London last month, the UK delayed its own AI Bill until at least next summer. The UK-US Tech Prosperity Deal was still being negotiated, and ministers were wary of making the UK unattractive to Silicon Valley tech giants. The biggest investment package of its kind in British history on the table, with 250 billion pounds set to flow both ways across the Atlantic.

In the Labour Party’s manifesto before last year’s general election, they pledged: “Labour will ensure the safe development and use of AI models by introducing binding regulation on the handful of companies developing the most powerful AI models and by banning the creation of sexually explicit deepfakes.”

Parliament has cracked down on sexually explicit deepfakes, ruling that predators who create them could face prosecution as the Keir Starmer bears down on vile online abuse. But the other half of the promise is still in limbo, with powerful AI companies so far getting off easy. Instead of prioritising safety and copyright issues, the government is dithering in favour of the likes of Nvidia, Microsoft, OpenAI and the pockets of BlackRock.

[ 👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io]

Now, the UK is falling behind its European neighbours. The sheer velocity of AI development is outrunning policymakers, and legislation may even be outdated when it finally does take effect. Italy passed a set of laws in September, limiting child access to AI and imposing prison terms for damaging use, such as generating deepfakes. The EU AI Act entered into force last year, marking the world’s first comprehensive legislation. Across the Atlantic Ocean, on the other hand, regulation is fragmented, and prospects for broad Congress-passed legislation remain doubtful.

Just last Thursday, a “bold” step was taken in the right direction: a new AI Sovereignty deal was struck between the Ministry of Justice and OpenAI where the tech giant will enable its business customers to store their data on British soil for the first time. British businesses will be able to host data on more secure, sovereign servers to reinforce national resilience in the face of growing global cyber threats.

But bolder steps are possible. In the absence of legislation, AI companies must take matters into their own hands. The industry should take the lead on self-governance and ethical guardrails to build public trust and pre-empt potentially heavy-handed or poorly informed regulation.

Centralised AI systems may offer scale, but they often fall short on trust and adaptability. That’s why decentralised AI, through approaches like blockchain-powered federated learning, is so vital. It lets countries and organisations work together, share insights, while keeping their data secured locally without putting all the control in one place.

Tech companies must commit to greater model transparency, accountability and data sovereignty. Blockchain is a secure and dependable way to achieve all of the above.

Transparency

The black box nature of powerful AI models is a major driver of public distrust. True self-governance requires tech companies to move beyond vague ethical statements and commit to radical model transparency. This means not just releasing a policy document, but openly documenting training datasets and error rates.

Being explicit about the datasets used to train models, including their origin, potential biases, and limitations. If a model’s training data disproportionately represents one demographic, the company must proactively disclose this, acknowledging the predictable bias in its outputs.

Publishing accessible documentation on what a model can and cannot do, including its error rates across different demographics, is fundamental to building trust. 

Accountability 

When an AI system causes harm – be it a biased loan algorithm or an inaccurate medical diagnosis – the public needs a clear chain of responsibility, not a corporate shrug. Accountability means building systems that are designed to be auditable by a third party.

Data sovereignty for countries and users

Perhaps the most critical principle of data sovereignty. As a bare minimum, this is the requirement that data is subject to the laws and governance of the country where it is collected or stored. The recent AI Sovereignty deal between the UK Ministry of Justice and OpenAI is a tentative step in the right direction, but it isn’t bold enough. It reinforces the centralised risk by consolidating data with a single, foreign-controlled tech giant.

We must take it a step further and pursue not just AI sovereignty but user sovereignty, giving individuals and data contributors more rights than governments currently demand.

This fundamental power imbalance can only be rectified by building an entirely new infrastructure: Decentralised AI (DeAI). This is where the combination of Federated Learning (FL) and Blockchain Technology becomes essential. FL solves the data movement problem by keeping data localised, and the blockchain solves the trust and control problem by removing the centralised coordinating entity.

Read our recent blog to find out why blockchain isn’t a cliché at FLock.io, it’s essential for scaling federated learning. 

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer. We just launched our API Platform.

Find out more about FLock.io by reading our docs. See a sneak peek of new launches coming in Q3/Q4 2025.

For future updates, follow FLock.io on X.

Head over to FLock’s blog for this article and other AI deep dives: https://www.flock.io/blog/tech-companies-shouldnt-wait-for-policymakers-to-regulate-ai

The answer is: definitely not. Blockchain is essential to our delivery of collaborative but secure models, and it has little to do with crypto cliches. It’s so paramount that we’re named after it: FLock stands for Federated Learning on Blockchain. Here’s why decentralised FL (deFL) is infinitely more scalable and secure than centralised FL (ceFL). 

The concept of federated learning (FL) was invented by Google in 2016. Instead of collecting vast amounts of sensitive data into a single, central location, federated learning lets multiple entities collaboratively train a model by bringing the training process directly to the data – then just collecting the updates, not the raw data.

👋 Hi there! If you’re here to find out more about FLock.io, follow us on X, read our docs and sign up to AI Arena on train.flock.io

The weaknesses of ceFL: a selfish central server, security risks and biased model output

Traditional FL is presented as a cure-all for what centralised AI comes under fire for – but in reality, the ceFL of today is riddled with weaknesses, bottlenecks and ethical question marks.

CeFL is controlled by tech giants like Apple and Google, which are plagued by a black box problem of opaque decision-making. Trust in such corporations is faltering as they put profit margins before public good, and exclude the community in determining the direction of AI development. Updates are sent to a central server, which monetises everything and grants no ownership to contributing nodes.

Problematically, it is in fact technologically viable to expose the base ceFL node’s data. Just like with other kinds of centralised AI, there is also a single point of failure which carries hazards for data privacy and security.

In ceFL, the model output can end up being biased because the central server has the final say. A central server coordinates the entire training process, dictates which nodes can join, and determines the validation set (the “correct” outcome).

This is dangerous because the central server – not the collective participants – has the sole authority to decide if an aggregated model is an improvement. This can result in biased models aligned with the central server's objectives, a problem exacerbated by the fact that fewer people are allowed to join as training nodes.

In a cartoon strip, Google AI presents FL as a privacy-preserving technique that can win back a tech company’s users at a time when their “brand is in shambles” from a “privacy nightmare”. 

The best data lives right here at the source: on the devices we use every day. FL lets us train a smarter centralised model on decentralised data, without ever collecting your sensitive data. It trains on your data in just a few minutes, then sends its training results (not your raw data) to the central server.

In essence, a central supervising node orchestrates the collaborative training of a model across multiple client devices. The server distributes a global model, and clients train it locally on each device before sending model updates back to the server which aggregates them to create an improved model. The updates are encrypted, and secure aggregation enables the server to combine the encrypted results, and only decrypt the aggregate.

But as the cartoon confesses, a core security weakness remains with ceFL: it’s technologically possible for the server to access raw data from the individual nodes. In a ceFL system, the training nodes are isolated from each other but can be exposed to the central authority. While this is useful for catching out malicious actors, it creates a significant privacy risk.

The existence of such a weakness requires participants to simply trust the central entity for security. This prevents high-value data silos (e.g., competing banks, hospitals) from ever joining, severely limiting the quality and diversity of the model as the network cannot grow.

A common application of ceFL is the keyboard autocompletion feature. The danger is clear: the company managing the server could potentially see everything you type.

FLock’s deFL blockchain solution: decentralisation, validation by consensus and incentive mechanism tokenomics

That’s where blockchain comes in. Blockchain is a core security and decentralisation tool that closes a whole array of loopholes. It solves systemic flaws inherent in traditional federated learning, making it more advanced and scalable. Unlike ceFL, deFL avoids using a central server to coordinate training and aggregate the updates. Instead, clients communicate model updates with their neighbour, decentralising the supervising node.

FLock.io uses blockchain to both protect user privacy and prevent malicious activity through a system of crypto incentives. We have enough mechanisms with rewards and penalties to form our very own sophisticated tokenomics. Read our blog about crypto incentives in deAI to find out more about the difference between crypto incentives in deAI and cryptocurrencies, and how such tokens drive participation.

Blockchain is powerful at mitigating malicious trainers, stopping them from carrying out “lookup attacks” to poison the FL model. It enforces a “Proof of Quality” consensus, secured by clever incentives. The specific problem FLock solves is this: if you have meaningful data, you should not need permission to participate in FL. We use algorithms and the blockchain to decide publicly, transparently, and democratically whose data qualifies for training and whose is unqualified and should be discarded. Contributors are incentivised, while those providing poor data are slashed. Because this involves economic incentives and penalties, a decentralised consensus mechanism is essential – no single entity can have the final say.

In our system, FL training nodes are randomly assigned as proposers and voters across multiple training rounds. These randomised voters validate new model proposals using their own local data, ensuring model quality. Only a model confirmed to be better is aggregated and adopted as the new base model for subsequent rounds. If malicious actors are identified, their staked $FLOCK tokens are slashed (i.e., forfeited). Conversely, if a model is accepted for aggregation, the proposers are rewarded with $FLOCK tokens and other incentives.

FLock's deFL cycle

Phase 1: registration. Stake $FLOCK to register as an FL node.

Phase 2: random assignment. Randomly assign the proposers and validators.

Phase 3: local training & model aggregation. Aggregate local model parameters into a new model.

Phase 4: validation & voting. Validate new model with local data. Voting results are recorded on blockchain, determining if the new model is better than the consensus base model.

Phase 5: reward proposers & consensus. Reward proposers and replace base model for next round of FL (or slash tokens and restart if the model fails).

Key differences between ceFL and FLock’s blockchain-powered deFL are in this table:

In summary, the key differences that FLock’s blockchain-powered deFL brings include: 

• A decentralised peer-to-peer network instead of a hub-spoke model with a central server. 

• Control is distributed among participants rather than a single entity controlling the training process. 

• Replacing the central server to orchestrate the training are blockchain and smart contracts. 

• Trust is distributed and cryptographically secured, not placed in a central server. 

• It’s more resilient to attacks due to decentralisation and no single point of failure or attack, and it has community-driven, on-chain governance.

To be clear, FLock.io is not trying to push the entire world on-chain. Our solution involves leveraging ceAI efficiency as well as deFL equality, without sacrificing the speed and bandwidth of ceAI. We’re excited to see how the whole ecosystem grows over the coming years.

The scalability of FLock’s deFL in action: the network effect

FLock’s deFL doesn’t just solve immediate technical problems. It unlocks the door to massive scalability through network effects. Traditional FL struggles to scale because its centralised nature demands a high degree of pre-existing trust between organisations, limiting participation to small, siloed groups.

Domain-specific model use cases, powered by blockchain consensus and incentives, will drive the mass adoption of FL. The more users adopt FL, the faster it will get. It’s comparable to how with social media, the more people make profiles, the better the user experience. 

Our technology is crucial for industries with heightened data sensitivity and fiduciary duties, including healthcare, pharmaceuticals, finance, government, and technology. Organisations in these sectors often find centralized AI products like ChatGPT too risky. They face strict compliance regulations, such as GDPR and HIPAA, and cannot risk exposing trade secrets or confidential client data.

Investment banking provides a clear example of these risks. Banks have strict policies prohibiting any tool or method that could share client data. A centralised AI presents obvious dangers: an analyst might use a tool like ChatGPT to grammar-check a draft registration statement or periodic report, potentially exposing confidential financial information that could be used to front-run the market if observed by OpenAI. Similarly, communications between investment banking and trading desks could leak information that facilitates insider trading.

deFL solves these problems. By deploying models on local devices, all training occurs locally. Only the model’s updated parameters – not the raw data – are ever shared. This allows everyone to benefit from a smarter, collectively trained domain-specific model without ever exposing their sensitive information.

Ultimately, the goal is to ensure the final model accurately represents the consensus of all FL training nodes, not that of the central server. This system also grants ownership percentages of the resulting model, calculated based on the $FLOCK tokens held by each node post-training. Therefore, large conglomerates cannot steal the model built from the nodes’ contributions.

We are proud to work with a range of leading clients and partners, including but not limited to the United Nations Development Programme, the Hong Kong Generative AI R&D Centre, Qwen by Alibaba Cloud, University College London Hospitals, and Aigen Sciences.

More about FLock.io

FLock’s ecosystem consists of three key components: AI Arena, a platform for competitive model training; FL Alliance, a privacy-focused collaboration framework that enhances models while preserving data sovereignty; and Moonbase, our new rewards layer. We just launched our API Platform.

Find out more about FLock.io by reading our docs. See a sneak peak of new launches coming in Q3/Q4 2025.

For future updates, follow FLock.io on X.