0x20 Web3 Security Bulletin

Crypto and web3 security insights, including tools, hacks, and regulations.

Insightful

Balancer Hardens V3 With Certora Re-Audit and New Guardrails

Balancer details how a four-year-undetected CSP exploit on V2 pushed the team to double down on preventive security in V3, which was unaffected by the incident. A fresh Certora re-audit reported no vulnerabilities, while new guardrails—minimum token balances, safer rounding, stable-pool imbalance caps, and flash-swap limits—aim to eliminate entire exploit classes before they emerge. (Balancer on X.com)

The Hidden Threats of Web2 Vulnerabilities in Web3 Systems

BlockApex explains how Web3 security often focuses on smart contracts while the Web2 layer—APIs, frontends, backends—stays under-audited. The post walks through web cache poisoning, IDOR, and SSRF in dApps and DeFi, with real incidents and hands-on labs. A few of the mitigation strategies covered: configure cache rules properly - if needed at all, include critical headers in cache keys, and always remember to validate untrusted user input. (BlockApex)

Is Their An Agent Moat Or Will Frontier Models Rule?

Bridging the gap between the web2 discussion above, @cmichel's commentary on Mario Zechner's post What I learned building an opinionated and minimal coding agent and 0xaudron's points below on the state of agent auditing vs. humans. I've been testing out agentic pen-testing tools and I wired up Hexstrike-AI MCP paired with a Kali Linux install. My first target was Natas' web security CTF. I fired off a prompt in Cursor to begin the test, but somehow Cursor didn't figure out I wanted to use the Hexstrike-AI MCP, I was about to shut it down and start over, but Cursor's agent set to the Auto model ripped through the whole CTF in no time. Afterwards I tried again with the correct prompt to engage the MCP server and Hexstrike got stuck on level 29. Anecdotal, and more testing is needed, but I found it interesting.

Sample Finding From My Cursor vs. Natas' CTF Testing

Level 32 → Level 33

Security Issue Discovered and Root Cause:

The application may have had issues with code injection through eval(), assert(), or similar functions where user input is executed as code. Alternatively, there may have been issues with template injection, where user input in templates is executed, or other code execution vulnerabilities.

// Code injection vulnerability
<?php
// User input executed as code
eval("echo " . $_GET['input'] . ";");
// Or template injection
$template = "Hello " . $_GET['name'];
eval($template);
// Allows code execution
?>

Risk - Why It Matters and What Could an Attacker Do:

Code injection allows attackers to execute arbitrary code on the server, leading to complete server compromise. Attackers can read files, execute commands, access databases, install backdoors, or use the server for further attacks. This is one of the most critical vulnerabilities, allowing complete system takeover.

Mitigation - How to Fix the Issue or Limit the Impact:

Never use eval(), assert(), or similar functions with user input. Avoid dynamic code execution entirely. If templates are necessary, use safe template engines that don't allow code execution. Implement proper input validation and output encoding. Use whitelist-based validation.

Can Your AI Audit Agent Do This Security Research?

0xaudron asks whether current AI audit agents can do the kind of security research he links in his thread. So far they can’t; he’s betting that changes soon.

SolidityGuard: Multi-Tool Smart Contract Auditor

Alt Research presents SolidityGuard, an advanced Solidity/EVM security auditor combining 104 curated vulnerability patterns with eight integrated tools including Slither, Mythril, Echidna, Medusa, Foundry, Halmos, and Certora. It offers web, CLI, Docker, and desktop surfaces, runs a 7-phase deep audit pipeline, aligns with OWASP Smart Contract Top 10 2025, and reports 100% detection across 85 CTF-style challenges. (SolidityGuard on Github)

Awal: agentic wallet for AI

0xErik.eth introduced awal, an agentic wallet for AI agents that can send, receive, pay, and trade. Available via npx; works with Coinbase agentic-wallet-skills for coding agents. (0xErik.eth on X.com)

Elliptic Outlines 2026’s AI-Driven Regulatory Future

Elliptic’s policy team sketches how regulators and compliance teams will lean on AI-enhanced blockchain analytics, data fabrics, and direct access to structured on-chain intelligence in 2026. The piece argues this tooling will sharpen AML/CFT efforts, improve resource allocation, and help both public and private sectors prioritize the highest-risk crypto activity. (Elliptic)

Companies in the news

Quiet week, Chainalysis continues to expand their product line. The rest of web3 must be in build mode.

Chainalysis’ Hexagate Backs MegaETH With Real-Time Smart Contract Defense

Chainalysis Team announces that Hexagate now protects the MegaETH ecosystem, offering real-time detection for smart contract exploits and abnormal on-chain behavior. The integration aims to give builders automated, low-latency security signal coverage so they can respond to threats before major funds are lost. (Chanalysis)

Chainalysis Expands Stellar Coverage With Automatic Token Support

Chainalysis Team highlights upgraded support for the Stellar network, extending coverage beyond XLM to automatic tracking of new fungible and NFT assets under Stellar’s token standards. Customers can now monitor these tokens in KYT, screening, and Reactor, enabling real-time risk alerts and investigations across Stellar’s growing tokenized ecosystem. (Chanalysis)

Gimme the loot

Surprisingly, no major hacks this week either. Is everyone just heads down building out OpenClaw tooling?

SlowMist stats this week

Total 2026 hack events: 22

The total amount of money lost by blockchain hackers is about $104,929,400

We must have regulations

DOJ Fines Paxful $4 million

The Block reports that Paxful Holdings was sentenced to pay $4 million after prosecutors said the platform profited from weak AML controls that let criminals move funds tied to fraud, prostitution, and sex trafficking. The DOJ said Paxful knew customers were moving crime proceeds and that founders called the resulting growth the “Backpage Effect.” About $17 million in bitcoin went to Backpage and a similar site from 2015 to 2022; Paxful earned at least $2.7 million. The company had pleaded guilty; the DOJ said it could not pay more than $4 million. (The Block)

CLARITY Act Stalls...

Commentators report that the U.S. CLARITY Act, a major digital-asset bill allocating most token oversight to the CFTC, had its January 14, 2026 House committee markup abruptly delayed with no new date set. The pause leaves exchanges, token issuers, and DeFi platforms in limbo as they await clearer federal rules on asset classification, market structure, and compliance expectations.

VCs & funding

Redaction! - CertiK Denies Immediate IPO Plans Despite Market Speculation

CertiK CEO Ronghui Gu clarified in a February 11, 2026 CoinDesk interview that the blockchain security firm has no concrete IPO timeline, walking back earlier speculation from January's World Economic Forum in Davos. While Gu described going public as a "natural next step" and long-term goal—potentially making CertiK the first publicly traded Web3 cybersecurity company at a $2 billion valuation—he emphasized no specific plan exists. The company continues focusing on trust recovery following 2025 controversies involving its Huione ecosystem audits. (CoinDesk)

Research corner

Code Is Not Law

Reyes, Tosato, and Hinkes (SSRN, Feb 2026) argue that ownership in digital assets comes from law, not from code. Using H.L.A. Hart’s legal theory, they show code has legal force only when positive law grants it—via legislation or private ordering. Otherwise code is a structural constraint without normative force. The paper applies the framework to smart contracts, DAOs, UCC Article 12, and to conflicts that recur with new technology. (SSRN)

And Just in Case You Missed It...

Code Is Law, a documentary exploring the legal boundaries of DeFi exploits through the story of teenage math prodigy Andean Medjedovic. The film follows cyber vigilantes hunting Medjedovic, who allegedly stole $64 million across protocols like Indexed Finance before fleeing justice, effectively testing whether smart contract exploits constitute theft or legitimate arbitrage in a court of law.

Play Video

LLM-Powered RiskTagger Automates Web3 AML Labeling

The authors present RiskTagger, an LLM-based agent that automates labeling of crypto laundering behavior in Web3 by replacing or augmenting human annotators. It extracts clues from unstructured reports, traces laundering across multiple chains, and generates auditor-ready explanations via a multi-module pipeline (clue extractor, multichain fetcher/reasoner, and explainer). On the Bybit hack case, RiskTagger reaches perfect clue extraction, strong expert-level consistency, and broad explanatory coverage. (arXiv)