Anthropic has developed Claude Mythos Preview, a powerful new AI model it claims is too dangerous for public release. The company announced on Tuesday it would limit access to a consortium of 40 tech firms including Apple, Amazon, and Microsoft through Project Glasswing, while withholding broader distribution due to security concerns.

Unlike its 2019 GPT-2 precedent, Anthropic's new model demonstrates autonomous capability to discover and exploit zero-day vulnerabilities in critical software. The company reports the system has already identified thousands of bugs across major operating systems and browsers, including a 27-year-old flaw in OpenBSD that evaded detection despite millions of automated scans. Anthropic is committing up to $100 million in Claude usage credits to the initiative.

Cybersecurity experts warn this represents a fundamental shift in the threat landscape. Anthropic's chief science officer Jared Kaplan described the model as potentially triggering an industry-wide reckoning over software security. The company projects annual revenue exceeding $30 billion in 2026, largely driven by Claude's popularity among programmers. As AI capabilities advance, Anthropic and industry leaders like Palo Alto Networks' Nikesh Arora warn that what once required months of human effort can now be accomplished in minutes by automated systems.