Anthropic’s Mythos Model Raises DeFi Security Concerns

Most DeFi smart contracts use publicly available code, allowing assets to be moved quickly once a vulnerability is exploited. Anthropic unveiled Claude Mythos Preview on April 7, 2026, touting its ability to identify and chain together zero-day vulnerabilities across large codebases. The same capability could strengthen defenses while also sharply lowering barriers for attackers.

Anthropic said on May 22 that about 50 Project Glasswing partners had identified more than 10,000 high-risk or critical vulnerabilities. On June 9, it released Mythos 5 exclusively to vetted partners. The US Treasury and Federal Reserve convened major bank CEOs for closed-door talks on April 7. The latest analysis, published on June 29, found that while AI can accelerate scanning, successfully stealing assets still requires human involvement.