<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/">
    <channel>
        <title>glassbox.foundation</title>
        <link>https://paragraph.com/@glassboxfoundation</link>
        <description>undefined</description>
        <lastBuildDate>Thu, 19 Nov 2026 11:41:26 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>https://github.com/jpmonette/feed</generator>
        <language>en</language>
        <image>
            <title>glassbox.foundation</title>
            <url>https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/99254e2357700ce68bad1b6a92990c165ea3ea5fc9e53ca92e1f6df4758ac703.jpg</url>
            <link>https://paragraph.com/@glassboxfoundation</link>
        </image>
        <copyright>All rights reserved</copyright>
        <item>
            <title><![CDATA[OpenAI Got the Diagnosis Right. The Prescription Is Missing a Layer.]]></title>
            <link>https://paragraph.com/@glassboxfoundation/openai-industrial-policy-missing-layer</link>
            <guid>zAOgfGdh7TZN2cJ5nAqP</guid>
            <pubDate>Wed, 08 Apr 2026 16:49:56 GMT</pubDate>
            <description><![CDATA[A Glassbox Foundation response to "Industrial Policy for the Intelligence Age" On April 6, OpenAI published a 13-page document proposing industrial policy for the age of superintelligence. Public Wealth Funds. Robot taxes. Four-day work weeks. Democratic oversight of alignment. Portable benefits decoupled from employers. Read it carefully. The diagnosis is sharp. The prescription operates entirely from the top down. And that's the problem. What they got right OpenAI names the core tension hon...]]></description>
            <content:encoded><![CDATA[<p><em>A Glassbox Foundation response to &quot;Industrial Policy for the Intelligence Age&quot;</em></p><hr><p>On April 6, OpenAI published a 13-page document proposing industrial policy for the age of superintelligence. Public Wealth Funds. Robot taxes. Four-day work weeks. Democratic oversight of alignment. Portable benefits decoupled from employers.</p><p>Read it carefully. The diagnosis is sharp. The prescription operates entirely from the top down. And that&apos;s the problem.</p><h2 id="h-what-they-got-right" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">What they got right</h2><p>OpenAI names the core tension honestly: AI concentrates power unless institutions intervene. They write it plainly - &quot;economic gains concentrate within a small number of firms like OpenAI.&quot; A company worth $300 billion publishing that sentence deserves credit.</p><p>The paper proposes three things that matter:</p><p><strong>A Right to AI</strong> - foundational access to models, not locked behind corporate subscriptions.</p><p><strong>Mechanisms for public input</strong> - so alignment isn&apos;t defined by engineers behind closed doors.</p><p><strong>An AI trust stack</strong> - provenance, verification, accountability for what models do in the real world.</p><p>These are the right building blocks. The question is: where do you put the foundation?</p><h2 id="h-what-they-missed" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">What they missed</h2><p>The entire document operates on one assumption: data flows to AI companies, and policy redistributes the benefits afterward.</p><p>Public Wealth Fund? Government collects money from AI firms, invests it, distributes returns. The data still left. The person still lost control. The fund compensates for extraction - it doesn&apos;t prevent it.</p><p>Portable benefits? Good idea. But benefits tied to what? To employment in an AI-driven economy where the real asset is your behavioral data, your health records, your decision patterns - not your labor contract.</p><p>Democratic oversight of alignment? Through what mechanism? The paper proposes &quot;representative input processes&quot; and &quot;model specifications.&quot; Who writes those specifications? Who enforces them? With what leverage?</p><p>The missing layer is data infrastructure. Not data policy. Infrastructure.</p><h2 id="h-the-difference-between-redistribution-and-architecture" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">The difference between redistribution and architecture</h2><p>OpenAI&apos;s framework redistributes value after aggregation. Collect the data. Train the model. Tax the profits. Distribute the returns. The person who generated the data is three steps removed from the value they created.</p><p>Glassbox starts from the opposite end.</p><p><strong>Data stays distributed.</strong> Your health metrics, financial decisions, daily habits, relationships - they live on your own server. Not in a cloud someone else controls. Not in a data lake feeding a foundation model. On your hardware, under your encryption keys.</p><p><strong>AI comes to the data.</strong> Compute-to-Data: models travel to user data inside attested secure enclaves. They learn from it. They leave. Raw data never moves. This isn&apos;t a policy proposal. It&apos;s a deployed architecture.</p><p><strong>Governance belongs to data owners.</strong> Not to corporate boards. Not to government-appointed panels. Expert Guilds - verified professionals in medicine, law, ethics - set domain-specific standards. Users hold veto power over changes affecting their data. Governance runs on quadratic voting. One person&apos;s voice matters regardless of wealth or token holdings.</p><p><strong>Access is conditional and revocable.</strong> Companies must pass independent security audits, publish governance reports, and declare prohibited uses before getting certified. Violate the terms - kill switch revokes access. Permanently.</p><p>This isn&apos;t redistribution. It&apos;s structural prevention. The concentration problem doesn&apos;t get solved by taxing concentrated profits. It gets solved by never allowing the aggregation.</p><h2 id="h-what-exists-today" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">What exists today</h2><p>ACUS Pod is an open-source personal AI server (AGPL-3.0). One instance has run since early 2026 - real health data, financial tracking, relationship mapping, habit systems. 15 AI agents managing 8 life domains. Not a prototype. Not a roadmap slide. A system processing real data daily.</p><p>Glassbox Foundation is a registered DAO LLC with a governance framework, certification process, and economic model. Health DataDAO deployed on testnet. Export pipeline operational.</p><p>The manifesto is published at glassbox.foundation. The code is at github.com/glassbox-foundation/acus-pod.</p><h2 id="h-where-we-converge" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Where we converge</h2><p>OpenAI wants democratic input on alignment. We built the governance mechanism.</p><p>OpenAI wants a trust stack with accountability. We built Compute-to-Data with kill switches.</p><p>OpenAI wants portable infrastructure not tied to employers. We built the personal AI server.</p><p>OpenAI wants shared prosperity from AI. We think prosperity means you own the asset - your data - not that a fund sends you a check.</p><p>The gap between their policy paper and implementation is exactly the layer we&apos;re building. From data up, not from policy down.</p><h2 id="h-an-invitation" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">An invitation</h2><p>OpenAI opened the conversation. They&apos;re offering fellowships and research grants for work that builds on these ideas. We&apos;re interested - not because we need validation, but because the architectures are complementary.</p><p>Policy without infrastructure is aspiration. Infrastructure without policy is plumbing. The Intelligence Age needs both.</p><p>We emailed their team today. We&apos;ll share what happens.</p><p>If you believe data sovereignty is the missing piece of AI industrial policy - read the manifesto. Run a Pod. Join the governance.</p><p>The transition to superintelligence will be governed. The question is whether you&apos;re at the table or on the menu.</p><hr><p><em>Paweł Wyszomirski</em> <em>Glassbox Foundation</em> <em>glassbox.foundation</em></p>]]></content:encoded>
            <author>glassboxfoundation@newsletter.paragraph.com (Paweł Wyszomirski)</author>
            <category>governance</category>
            <category>data-sovereignty</category>
            <category>ai-policy</category>
            <enclosure url="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/d7fe541de9337867d91dcf91cb2f0807ea4531865f19acc75658a7c32bfb598d.jpg" length="0" type="image/jpg"/>
        </item>
        <item>
            <title><![CDATA[The Glassbox Manifesto]]></title>
            <link>https://paragraph.com/@glassboxfoundation/manifesto</link>
            <guid>EL2mwjWZulC5zldB8mB7</guid>
            <pubDate>Sun, 01 Mar 2026 15:37:52 GMT</pubDate>
            <description><![CDATA[The Glassbox ManifestoYour Data. Your Life. Your Terms.ACUS - a personal AI pod under your control. Glassbox - certification + licensing + settlement. Compute-to-Data - models come to the data. Data doesn't leave.I. The Broken PromiseThe internet was supposed to empower individuals. Instead, it built the most efficient extraction machine in history. Every heartbeat tracked by your watch, every photo of your child, every late-night search about your health - captured, packaged, and sold. Not t...]]></description>
            <content:encoded><![CDATA[<h1 id="h-the-glassbox-manifesto" class="text-4xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">The Glassbox Manifesto</h1><h2 id="h-your-data-your-life-your-terms" class="text-3xl font-header !mt-8 !mb-4 first:!mt-0 first:!mb-0">Your Data. Your Life. Your Terms.</h2><blockquote><p><strong>ACUS</strong> - a personal AI pod under your control. <strong>Glassbox</strong> - certification + licensing + settlement. <strong>Compute-to-Data</strong> - models come to the data. Data doesn't leave.</p></blockquote><hr><h3 id="h-i-the-broken-promise" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">I. The Broken Promise</h3><p>The internet was supposed to empower individuals. Instead, it built the most efficient extraction machine in history.</p><p>Every heartbeat tracked by your watch, every photo of your child, every late-night search about your health - captured, packaged, and sold. Not to serve you. To train machines you have no say over.</p><p>The largest AI companies in the world are built on human lives. They collected data under vague consent. They profited without sharing. And now they're running out of public data to scrape - so they're coming for what's left: the private, intimate details of human existence.</p><p>This is not a technology problem. This is an ownership problem.</p><hr><h3 id="h-ii-the-simple-truth" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">II. The Simple Truth</h3><p><strong>You are the most valuable dataset that has ever existed.</strong></p><p>No synthetic data can replace the real texture of a human life - your decisions, your relationships, your mistakes, your growth. From the moment you were born, you began generating information that is uniquely, irreplaceably yours.</p><p>But right now, you don't control it. Corporations do. Platforms do. Algorithms do. You are a passenger in your own story.</p><p>We believe this must change.</p><hr><h3 id="h-iii-the-vision" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">III. The Vision</h3><p>Imagine a world where every person, from birth, has a personal intelligence system.</p><p>Not a social media profile. Not a cloud account someone else controls. A sovereign space - your own AI that understands your life, protects your data, and works for you.</p><ul><li><p>It monitors your health to <strong>reduce preventable risk and surface patterns early</strong> - not to sell you pills.</p></li><li><p>It tracks your finances to <strong>build your freedom</strong> - not to push you products.</p></li><li><p>It remembers your relationships, your values, your goals - because they matter to <strong>you</strong>, not to an advertiser.</p></li><li><p>It organizes your habits, decisions, and daily plans into a system that learns <strong>with</strong> you, not about you.</p></li></ul><p><strong>This system already exists. We are building it.</strong></p><p>ACUS is a personal AI server that collects, processes, and protects the data of your life. It runs on your terms. The data stays under your control.</p><p>But personal sovereignty is not enough. In a world where AI shapes everything - from what you see to what you believe - you also need a voice in how AI behaves. Not as a consumer. As a citizen.</p><hr><h3 id="h-iv-the-bridge" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">IV. The Bridge</h3><p>This is where Glassbox Foundation enters.</p><p>To understand the architecture, think of it in three parts:</p><ul><li><p><strong>ACUS</strong> = your device, your server. It holds your data and runs your personal AI.</p></li><li><p><strong>Glassbox</strong> = the standards body. It manages certification, licensing policies, and settlement between data owners and AI companies.</p></li><li><p><strong>Compute-to-Data</strong> = the method. AI models travel to your data, learn from it, and leave. Workloads are signed and run inside attested secure enclaves with least-privilege access. Your raw data never moves.</p></li></ul><p>Glassbox never stores your personal data. It brokers policies, certification, and settlement - while computation happens where your data lives.</p><p>Here is what Glassbox ensures:</p><ol><li><p><strong>Only certified AI gets access.</strong> Companies must pass certification before they can request data. Certification requirements include: independent security audit, a public model governance report, explicit prohibited-use declarations (no military, no surveillance without consent), an impact assessment, and a binding enforcement mechanism with financial penalties for violations.</p></li><li><p><strong>You set the rules.</strong> Through democratic governance, you participate in defining what values AI systems must follow. Expert guilds - verified professionals in medicine, law, ethics - set domain-specific standards. Users hold veto power over changes that affect their data categories. Governance uses quadratic voting and delegation to prevent plutocracy - one person's voice matters regardless of how many tokens they hold.</p></li><li><p><strong>You can get paid.</strong> When an AI model trains on your data, you receive a data dividend. Dividends depend on demand, data category, and your opt-in choices. There are no guaranteed returns - this is not a financial product. You may also choose to share nothing and owe no one an explanation.</p></li><li><p><strong>Your data stays home.</strong> Through Trusted Execution Environments and federated learning, models learn from your data without seeing it in raw form. The model travels. Your secrets don't.</p></li></ol><hr><h3 id="h-v-how-it-works" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">V. How It Works</h3><p>Five steps from installation to impact:</p><ol><li><p><strong>Install your ACUS pod.</strong> Set up your personal AI server - self-hosted or in a trusted cloud. Connect your health devices, calendar, financial tools. Your data begins accumulating under your control.</p></li><li><p><strong>Choose what to share.</strong> Opt in per data category: health, financial behavior, daily habits, relationships - or nothing at all. Zero-share is always an option. Each category has independent consent settings.</p></li><li><p><strong>Join the governance.</strong> Participate in the Glassbox DAO - vote on certification standards, delegate to expert guilds, or simply observe. Your participation shapes which AI companies qualify for access.</p></li><li><p><strong>AI companies apply.</strong> Certified companies submit data requests through Glassbox. The protocol checks their certification status, matches their needs with willing data owners, and issues a smart license.</p></li><li><p><strong>Compute-to-Data runs.</strong> The AI model enters your secure environment, trains on your data locally, and extracts only aggregated gradients. Your raw data never leaves. You receive your data dividend. The cycle repeats.</p></li></ol><hr><h3 id="h-vi-why-now" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">VI. Why Now</h3><p>Three forces are converging:</p><p><strong>AI hunger.</strong> Large language models have consumed most of the public internet. The next frontier is private human data - health records, financial behavior, personal journals, family histories. Someone will broker access to this data. The question is: who?</p><p><strong>Regulatory awakening.</strong> The EU AI Act, GDPR, and emerging data sovereignty laws are creating a legal framework that favors the individual. Governments are looking for infrastructure that makes personal data rights real, not theoretical.</p><p><strong>Technical readiness.</strong> Trusted Execution Environments, federated learning, and blockchain governance have matured enough to make this viable. The pipes exist. What's missing is the institution.</p><p>Glassbox is that institution.</p><hr><h3 id="h-vii-the-principles" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">VII. The Principles</h3><ol><li><p><strong>Data is birthright.</strong> Every human being should have practical control and enforceable rights over the data generated by their existence - regardless of how local law defines "ownership." This is not a feature. It is a foundation.</p></li><li><p><strong>Sovereignty before profit.</strong> No amount of money justifies losing control over your own information. Monetization is a right, not an obligation.</p></li><li><p><strong>Transparency is the price of entry.</strong> Any AI system seeking access to human data must be auditable - open governance reports, public prohibited-use lists, verifiable compliance. Black boxes do not get keys.</p></li><li><p><strong>Governance belongs to people.</strong> The rules that shape AI behavior should be written by the humans whose lives are affected - through expert guilds, community votes, and enforceable veto rights - not by the companies that profit from them.</p></li><li><p><strong>Simplicity is strength.</strong> You should not need to understand blockchain to own your data. The technology must be invisible. The benefit must be obvious.</p></li><li><p><strong>Start small, think long.</strong> We begin with one server, one person, one life made better. We end with a global infrastructure for human data sovereignty. Every journey between those points is walked one step at a time.</p></li></ol><hr><h3 id="h-viii-what-we-know-we-cant-solve-yet" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">VIII. What We Know We Can't Solve Yet</h3><p>No privacy technology is perfect. We believe honesty about limitations builds more trust than false promises.</p><p><strong>What we mitigate:</strong> Direct data exposure (through Compute-to-Data and TEEs), unauthorized access (through certification and smart licenses), misuse after training (through kill switches that revoke certification and block future data access). We can stop future access, but we can't retroactively remove what a model has already learned.</p><p><strong>What we can't fully eliminate today:</strong> Model inversion attacks (reconstructing training data from model outputs), membership inference (detecting whether a specific person's data was used), and side-channel leaks from execution environments. These are active research areas across the industry.</p><p><strong>Our commitment:</strong> All core infrastructure is open-source and subject to independent security audits. We run a bug bounty program. We publish threat model updates as the technology evolves. When we discover a new risk, we disclose it.</p><hr><h3 id="h-ix-what-exists-today" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">IX. What Exists Today</h3><p>This is not a whitepaper dream.</p><p>ACUS - the personal AI server - is live. It tracks health metrics, daily habits, financial decisions, and personal relationships for its first user. It integrates with calendars, fitness devices, and communication tools. It runs on a private server with encrypted backups. The dashboard is accessible, the data is structured, the system learns.</p><p>Glassbox Foundation is registered as a DAO LLC. Its governance framework, certification process, and economic model are documented. The first vertical - rare disease data licensing - is designed and ready for pilot.</p><p>The website is live at glassbox.foundation. Grant funding is secured through Arbitrum DAO.</p><p>Glassbox is designed to interoperate with existing data infrastructure - including networks like Vana - adding an ethical governance layer that the marketplace alone cannot provide.</p><p>We are not asking you to believe in a concept. We are asking you to join a system that is already running.</p><hr><h3 id="h-x-the-invitation" class="text-2xl font-header !mt-6 !mb-4 first:!mt-0 first:!mb-0">X. The Invitation</h3><p>If you believe that your data should work for you, not against you - you belong here.</p><p>If you believe that AI should serve humanity on humanity's terms - you belong here.</p><p>If you are a developer, a researcher, a patient, a parent, a founder, or simply a human being who wants to stop being the product - you belong here.</p><p><strong>Own your data. Shape the AI. Live better.</strong></p><p>This is the Glassbox promise.</p><hr><p><em>Glassbox Foundation - 2026</em> <a target="_blank" rel="noopener noreferrer nofollow ugc" class="dont-break-out" href="https://glassbox.foundation"><em>glassbox.foundation</em></a></p>]]></content:encoded>
            <author>glassboxfoundation@newsletter.paragraph.com (Paweł Wyszomirski)</author>
            <category>data-sovereignty</category>
            <category>ai-governance</category>
            <category>manifesto</category>
            <category>web3</category>
            <category>dao</category>
            <enclosure url="https://img.paragraph.com/cdn-cgi/image/format=auto/https://storage.googleapis.com/papyrus_images/4742d42ab1d4a4c82d18b06a5981f2142ed16e782cd14c877a135afde5cc15bc.jpg" length="0" type="image/jpg"/>
        </item>
    </channel>
</rss>