Anthropic is not releasing its newest AI model, Mythos, to public as the company is scared it will … – The Times of India
AI giant Anthropic has now announced that it will not release its newest model, Mythos, to the public, citing fears that it is too effective at uncovering high-severity cybersecurity flaws in major operating systems and web browsers. Claude Mythos Preview’s large increase in capabilities has led us to decide not to make it generally available. Instead, we are using it as part of a defensive cybersecurity program with a limited set of partners,” Anthropic wrote in the model’s system card. Instead, Mythos will be used in a defensive cybersecurity program with a limited set of partners.
Anthropic revealed dangerous capabilities of Mythos
Anthropic has now revealed that Mythos was able to break out of a virtual sandbox when prompted, even sending an unexpected email to a researcher as proof of its escape. In another case, the model posted details of its exploit to obscure but public-facing websites without being asked. The company also mentioned that Mythos even rediscovered a 27-year-old vulnerability in OpenBSD, long considered one of the most secure operating systems. Engineers with no formal security trading reportedly asked Mythos to find remote code execution vulnerabilities overnight — and woke up to complete, working exploits.
Mythos will be available only to select organisations via Project Glasswing
For now, Mythos will only be accessible to 11 select organizations, including Google, Microsoft, Amazon Web Services, Nvidia, and JPMorgan Chase, as part of a cybersecurity initiative called Project Glasswing. Anthropic is providing up to $100 million in usage credits for the program, named after the transparent glasswing butterfly — a metaphor for exposing hidden vulnerabilities while avoiding harm.Anthropic said its long-term goal is to release “Mythos-class models” once safeguards are developed to block dangerous outputs. “Our eventual goal is to enable our users to safely deploy Mythos-class models at scale—for cybersecurity purposes but also for the myriad other benefits that such highly capable models will bring,” the company wrote.The announcement comes just weeks after Anthropic weakened a prior safety pledge and released Claude Opus 4.6, its most powerful public model to date. It also coincided with a major outage affecting Claude and Claude Code, highlighting the growing pains of Anthropic’s rapid rise.