Anthropic has limited early access to its new AI model Claude Mythos to about 50 organisations, citing its strong ability to find and exploit software vulnerabilities. It says the model found thousands of flaws and generated attack code far beyond its previous models. Security expert Bruce Schneier said the disclosure resembles responsible release but lacks enough information, especially on error rates. He urged broader access, more data sharing and support for independent audits.

Anthropic has shown some companies a new AI model under development, Claude Mythos, and introduced Glasswing, a cybersecurity project involving more than 40 major tech companies. The move aims to reduce the risk of hackers misusing Mythos. Anthropic says Mythos has already found thousands of high-risk vulnerabilities across major operating systems and web browsers. Casey Newton and industry executives warned advanced reasoning and widening access could intensify threats and concentrate power.