Claude Mythos Preview

Claude Mythos Preview is Anthropic's most capable cybersecurity AI model, released on 8 April 2026 exclusively to twelve partner organisations through Project Glasswing. The model scored 83.1% on the CyberGym vulnerability reproduction benchmark, against 66.6% for Anthropic's previous best model, and autonomously identified thousands of zero-day vulnerabilities including a 27-year-old flaw in OpenBSD , performance that prompted Anthropic to withhold it from public release on safety grounds.

Tom's Hardware subsequently published a critical review arguing the "thousands of zero-days" claim rested on only 198 manual reviews and that many flagged vulnerabilities were in outdated, no-longer-active software. The challenge does not dispute the benchmark score but raises questions about the real-world severity of the vulnerabilities identified. The controversy is a recurring pattern in AI capability claims: headline figures that are technically accurate but require significant qualification in context.

The US government's reaction was swift. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell summoned the CEOs of five Major Banks to an emergency meeting specifically to discuss Mythos, marking the first time federal financial regulators have convened Wall Street leadership over a single AI model's capabilities. The model's restricted distribution, combined with $100M in usage credits for Glasswing partners, places Mythos at the intersection of AI capability competition, financial-sector cybersecurity, and the emerging debate over whether frontier AI systems require government oversight before deployment.