Crypto firms chase AI guard‑dog that may backfire

A handful of crypto platforms are reaching out to a new AI model called Claude Mythos, hoping it can help them spot hidden security holes. The tool is still in a limited release because its creators worry it could be used to find and exploit weaknesses on a massive scale.

• Crypto exchanges such as Coinbase have talked with the makers of Mythos, while Binance and a custody company called Fireblocks are already testing other Anthropic models for penetration tests.
• Those firms say the AI can uncover bugs that even skilled human testers miss.

Anthropic claims Mythos discovered long‑hidden flaws in older systems that had survived for almost thirty years. Yet the same power could let bad actors locate and attack digital vaults, a major concern for companies that hold billions in crypto assets.

The sector has faced several high‑profile breaches, including a data leak last year that exposed customer information. Because of these risks, Anthropic is cautious about who can use Mythos and has kept it away from most crypto companies.

Bottom line: The crypto community wants a powerful defensive tool but must balance that with the risk of giving attackers a similar advantage.