Anthropic released its latest AI model, Claude Mythos, this week, touting built-in safeguards. But the launch is already drawing sharp warnings from the crypto security world. Venture capitalist Simon Dedic said the new generation of Anthropic models reduces the cost and skill required to find crypto exploits to 'basically zero' — a claim that puts the industry on notice.
What Claude Mythos brings
The model comes with what Anthropic calls 'constitutional guardrails' meant to prevent abuse. The company has positioned Mythos as its most safety-conscious release yet. But the underlying capability is what has security researchers paying attention: the same reasoning power that helps developers write secure code can, in the wrong hands, find the holes in it.
Dedic’s warning
Simon Dedic, a venture capitalist focused on crypto infrastructure, didn't mince words. In a statement this week, he said Anthropic's latest models effectively democratize exploit discovery — making it cheap and accessible to anyone with malicious intent. He didn't elaborate on specific tests or examples, but the claim lands at a moment when crypto platforms are already struggling to patch vulnerabilities fast enough.
The exploit landscape
This isn't the first time AI has been flagged as a growing threat to blockchain security. But Dedic's framing — 'basically zero' cost and skill — suggests the bar is about to drop sharply. If accurate, it means small-time attackers could soon operate at a level previously reserved for elite hacking teams. The timing isn't great: the industry is still dealing with a backlog of unpatched smart contract bugs from the past year.
What comes next
Anthropic hasn't responded to Dedic's remarks. The safeguards in Claude Mythos are untested at scale against a motivated adversary. Whether they hold up — or whether the crypto sector needs to rush its own defensive AI tools — remains an open question with no deadline in sight.
