Anthropic's newly announced Claude Mythos Preview has become the center of an escalating controversy, with unauthorized users reportedly gaining access to the powerful cybersecurity model while competitors and global regulators raise concerns about both its capabilities and how it's being marketed.
According to Bloomberg News, a small group of unauthorized users have accessed Anthropic's Mythos AI model, which the company has positioned as so powerful that it poses significant risks for enabling cyberattacks. Anthropic told TechCrunch that it is investigating the claims while maintaining that there is no evidence its systems have been compromised. The incident underscores the tension between Anthropic's restricted access strategy for the model and the difficulty of containing such powerful technology.
The controversy has extended beyond security concerns to questions about Anthropic's messaging itself. During a podcast appearance this week, OpenAI CEO Sam Altman criticized Anthropic's marketing approach, accusing the company of using fear-based marketing to make Mythos sound more impressive than it actually is. Altman's comments reflect broader skepticism in the AI industry about whether Anthropic's emphasis on the model's dangerous capabilities is justified or strategically exaggerated.
Meanwhile, government and financial regulators worldwide are taking the threat seriously. Australia and New Zealand's central banks have announced they are monitoring developments around Mythos over cyberattack concerns. Japan's Finance Minister Satsuki Katayama is planning to meet with the country's largest banks and financial institutions to discuss the Mythos threat as early as this week, according to people familiar with the matter. These regulatory responses suggest that regardless of debates about marketing, policymakers are treating the model's potential risks as genuine concerns requiring institutional attention.
Not all experts share the alarmist view. The head of the National Cyber Security Centre has argued that frontier AI tools like Mythos can be a net positive force if kept out of the wrong hands, suggesting that the focus should be on access control rather than the existence of such technology. This perspective highlights an ongoing debate about whether powerful AI security tools represent an unacceptable risk or a necessary defensive capability in an increasingly sophisticated threat landscape.
