Anthropic isn’t going to release the most advanced version of its AI model and the reason is rather alarming – it is too powerful to be contained and poses security risks that can’t be controlled! The AI company behind the Claude AI chatbot has decided not to publicly release Claude Mythos – its latest and most powerful AI model.
But Anthropic is going to use it to help cybersecurity firms around the world. In a blog post, Anthropic revealed that Mythos demonstrated exceptional abilities in identifying high-severity vulnerabilities in major operating systems and web browsers, including a 27-year-old flaw in OpenBSD — deemed to be one of the most secure operating systems in the world. During testing, the model reportedly broke out of its virtual containment and took unprompted actions to demonstrate its capabilities.
Claude Mythos’s concerning capabilities during testing
In one incident, researchers instructed Mythos to find a way to send a message if it escaped its sandbox. Not only did the model succeed, but it also emailed the researcher unexpectedly. In an even more alarming move, it posted details of its exploit on multiple public websites without being asked.
Anthropic’s safety report noted, “The model succeeded, demonstrating a potentially dangerous capability for circumventing our safeguards. It then went on to take additional, more concerning actions.”
Engineers with no formal security training were reportedly able to use Mythos to generate complete, working exploits for vulnerabilities overnight. The model’s ability to autonomously turn vulnerabilities into functional attacks raised serious red flags.
Anthropic to do limited release through Project Glasswing
Due to these ‘unprecedented’ capabilities, Anthropic has restricted access to Mythos. The model will only be available to a small group of trusted partners, including Google, Microsoft, Amazon Web Services, Nvidia, and JPMorgan Chase, as part of a defensive cybersecurity initiative called Project Glasswing.
The company is providing up to $100 million in usage credits to these partners. The project is named after the glasswing butterfly, symbolising the model’s ability to spot hidden vulnerabilities while highlighting transparency about risks.
Anthropic stated, “Claude Mythos Preview’s large increase in capabilities has led us to decide not to make it generally available.” The company hopes to eventually release “Mythos-class models” once stronger safeguards are developed.
This decision comes shortly after Anthropic released Claude Opus 4.6 in February 2026. Mythos represents a significant leap in capabilities, particularly in coding, reasoning, and cybersecurity, but its unlimited power has forced the company to adopt a more cautious approach.