US Officials Warn Wall Street That Anthropic's Mythos AI Could Usher In a New Era of Cyber Risk
Federal regulators and senior officials, including Federal Reserve Chair Jerome Powell, privately cautioned bank CEOs about the potential dangers of Anthropic's restricted frontier model, as JPMorgan and Morgan Stanley were permitted to test it for defensive security purposes.
US government officials and financial regulators warned major Wall Street banks that Anthropic's newest AI model, Mythos, could introduce a new era of cybersecurity risk, Bloomberg reported on April 13. Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent were among officials who privately cautioned bank chief executives about the potential for advanced AI systems to amplify existing vulnerabilities in financial infrastructure, according to people familiar with the discussions.
The concerns centered on how a model of Mythos's capability could be misused to automate social engineering campaigns, identify weaknesses in banking networks at scale, or combine multiple lower-severity vulnerabilities into critical exploits. JPMorgan Chase and Morgan Stanley were among a small closed group of financial institutions that had been granted access to Mythos through Anthropic's restricted Project Glasswing initiative, where they were testing the model specifically to identify vulnerabilities and strengthen their own defenses.
Anthropic launched Project Glasswing in early April, initially restricting Mythos Preview to a consortium of eleven organizations including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. The company characterized the restricted rollout as a deliberate safety measure after internal testing found the model demonstrated extraordinary capabilities in uncovering zero-day vulnerabilities across major operating systems and browsers.
The Bank of England separately announced it was preparing its own assessment of the implications of Mythos for the UK banking sector, according to reporting cited in the Bloomberg coverage. Cybersecurity professionals noted that while models with similar capabilities were likely to emerge from other labs within months, the concentration of such tools in few hands created an asymmetric risk environment favoring well-resourced threat actors.
Read the original reporting at Bloomberg.
