Anthropic Withholds Powerful AI 'Mythos' Citing Cybersecurity Risks

Anthropic, a leading firm in artificial intelligence safety and research, has made the significant decision to limit the release of its latest model, Mythos, due to concerns about its powerful cybersecurity capabilities. The company announced it will not proceed with a wide public release after an internal red team discovered the model could be used to orchestrate complex cyberattacks.

While Mythos is a general-purpose AI, it possesses exceptionally advanced coding and reasoning skills. These abilities allowed it to identify long-dormant vulnerabilities in critical internet software, including a 27-year-old bug. However, these same strengths proved to be a double-edged sword. Anthropic's safety researchers demonstrated that Mythos could also be instructed to chain together exploits, effectively creating novel attacks against all major operating systems and web browsers.

Faced with this potential for misuse, Anthropic has opted for a cautious approach. Through an initiative named Project Glasswing, the company is providing early access to Mythos to a select group of trusted partners. This group includes tech giants like Apple, Amazon, Microsoft, and Google, as well as several specialized cybersecurity firms. The goal is for these organizations to use Mythos to proactively hunt for vulnerabilities in their own systems, with all findings being shared to develop stronger safety guardrails for future AI models.

This move highlights the growing dilemma in the AI industry: how to balance the rapid advancement of powerful tools with the critical need for responsible deployment and public safety. Anthropic's decision to keep Mythos under wraps for now marks a pivotal moment in the ongoing conversation about AI regulation and ethical development.