Anthropic restricted Mythos capabilities after warning advanced AI could discover vulnerabilities autonomously.
Security experts warn bridges, oracles, validators, and wallets face growing AI risks.
Claude Opus previously helped uncover a critical vulnerability within Zcash's Orchard protocol.
Anthropic, an artificial intelligence (AI) safety and research company known for developing Claude, has launched Claude Fable 5, its most powerful AI model now available to the public. Anthropic said the company removed some advanced cybersecurity features, which are capable of finding software vulnerabilities on its own.
For the crypto industry, that raises a bigger question, if AI can discover security flaws faster than humans, could crypto platforms and investors face greater risks from future cyberattacks?
Claude Fable 5: Most Powerful AI For Public
Claude Fable 5 belongs to Anthropic’s next-generation “Mythos-class” family of AI models, a category the company previously considered too powerful for general release.
According to Anthropic, Fable 5 outperforms all of its previous public models across most major benchmarks. It is designed to handle complex research, coding, and problem-solving tasks.
Meanwhile, Claude Fable 5 comes with strict limitations.
Likewise, when users ask questions related to cybersecurity attacks, biological threats, chemical risks, or model extraction techniques, the system automatically redirects those requests to Claude Opus 4.8, a less capable model.
The company said these safety measures are used in less than 5% of sessions and help keep the model safe while allowing public access.
Claude Mythos 5: Restricted Version for Public
Alongside Fable 5, Anthropic also introduced Claude Mythos 5, a more powerful version of the model. The company said it can find unknown software vulnerabilities on its own, including security weaknesses that developers and security teams may not have discovered yet
Access is limited to trusted organizations through Project Glasswing, a program developed alongside the U.S. government. Current users include Microsoft, Nvidia, Cisco, Google Cloud, and AWS.
Anthropic made its position clear in the launch announcement,
“No entity currently has sufficient safeguards for a full public release of Mythos-level models.”
This follows the company’s earlier Mythos Preview release in April 2026, which was also restricted to a small group of vetted cybersecurity organizations.
Why the Crypto Industry Should Worry
The crypto industry has a direct reason to watch this development. Anthropic revealed that Claude Opus 4.8 previously helped uncover a critical vulnerability in Zcash’s Orchard protocol.
That discovery was made using a model that is less capable than Mythos 5.
This means that, if a weaker model could find a serious flaw in a privacy-focused blockchain, what might a far more powerful Mythos-class model uncover?
AI Could Target Crypto Infrastructure
Many crypto teams focus heavily on smart contract audits. But Anthropic’s launch highlights a different concern.
Security experts say the biggest risks may not come from smart contracts. Instead, AI could find weaknesses in crypto systems like cross-chain bridges, oracle networks, and wallet security tools.
If an oracle is compromised, it can affect many DeFi platforms at once. Weaknesses in wallet security could also put billions of dollars in crypto funds at risk.
Lastly, Anthropic says models like Fable 5 and Mythos 5 have “the potential to do profound good for the world.”