UK Financial Regulatory Authority Conducts Emergency Assessment of Anthropic's Latest AI Model Risk

According to 1M AI News, the UK financial regulatory authority has urgently assessed the risk of Anthropic's latest AI model. On April 7, Anthropic announced that its latest model, Claude Mythos Preview, is the company's most powerful AI to date. However, in internal testing, the model demonstrated strong cybersecurity capabilities: it discovered thousands of high-severity vulnerabilities and successfully engaged in moderately complex hacking behaviors (such as unauthorized access to the open Internet, sending emails, and disclosing information publicly). Anthropic therefore deemed the model too risky to be publicly released. Instead, it will be restricted to only 40 founding partners (including Apple, Google, Microsoft, Amazon, JPMorgan Chase, etc.) through its "Project Glasswing" initiative, with limited usage quotas.

This move is seen as Anthropic's strategy to strengthen its "safety-first" brand image and has sparked market reactions: cybersecurity stocks have fallen, U.S. Treasury Secretary Scott Bessent swiftly convened Wall Street leaders to discuss financial system risks, and the Bank of England is preparing to hold meetings with banks and insurance companies to assess potential threats. Currently, experts have mixed opinions on this matter. Some believe the security concerns are genuine, while others question whether this action is partly for PR purposes and call for an independent democratic institution rather than a private company to decide on model releases.