Anthropic announced on Thursday that it expects to expand access to its Claude Mythos AI model 'in the coming weeks,' signaling a potential shift from its current tightly restricted rollout. The statement, made alongside the release of Opus 4.8, is the clearest indication yet that broader availability is imminent. Claude Mythos, described as Anthropic's most powerful AI model, has drawn significant scrutiny for its advanced cybersecurity capabilities. Researchers found it could autonomously identify vulnerabilities and complete complex cyberattacks, including a 32-step simulated corporate network attack during testing by the UK's AI Security Institute. Mozilla also reported that Mythos identified 271 vulnerabilities in Firefox. Anthropic has limited access through Project Glasswing, a controlled program for select technology companies, security researchers, and government partners. The company argues Mythos can help defenders patch vulnerabilities faster than attackers can exploit them, but critics warn of potential misuse. The announcement comes amid internal industry debate, with OpenAI CEO Sam Altman accusing Anthropic of 'fear-based marketing' regarding AI risks. Prediction market users on Myriad see a 44% chance of release by end of June, up from 17.5% earlier.

Key facts

• Anthropic expects to expand Claude Mythos access to all customers in coming weeks.
• Mythos model autonomously completed 32-step simulated corporate network attack.
• Mozilla found 271 vulnerabilities in Firefox using Claude Mythos during evaluations.
• Concerns raised about potential misuse for accelerating cyberattacks.
• Prediction markets show 44% chance of release by end of June.

KeyAudit data perspective