What happened: Anthropic has launched Claude Fable 5, a cutting-edge model from its "Mythos-class" family that outperforms previous versions in coding and complex problem-solving. However, to prevent misuse, the company has stripped the public version of its ability to autonomously find software vulnerabilities. For high-level cybersecurity tasks, Anthropic maintains a restricted version called Claude Mythos 5, which is only accessible to trusted organizations like Microsoft, Nvidia, and Google Cloud through its Project Glasswing program.
Why it matters: This development signals a growing threat to the crypto ecosystem. Since a less capable model previously helped uncover a vulnerability in Zcash's Orchard protocol, the arrival of Mythos-class AI suggests that hackers could soon use advanced AI to target critical infrastructure like cross-chain bridges, oracle networks, and digital wallets, potentially putting billions in assets at risk.
