Anthropic Launches Claude Fable 5, Raising Concerns Over AI-Powered DeFi Vulnerability Discovery

Anthropic has released Claude Fable 5, its first publicly available Mythos-class AI model. To limit potential misuse, the company has introduced an automatic safety system that redirects prompts involving cyberattacks, chemical or biological threats, and model distillation to the less capable Claude Opus 4.8.

The measure is particularly significant because Claude Opus 4.8 has already been used in advanced security research, including helping uncover a critical flaw in the Orchard protocol of the privacy-focused cryptocurrency Zcash.

The launch of Fable 5 has also renewed concerns about the growing role of frontier AI in cybersecurity. As AI capabilities advance, such models could speed up the detection of weaknesses in privacy coins, smart contracts, and DeFi platforms, enhancing security audits while potentially increasing risks for projects that have not undergone thorough code reviews.

Source: Anthropic