European regulators have expressed concern over being "sidelined" regarding Anthropic’s unreleased AI model, Claude Mythos, which possesses advanced "super-hacking" capabilities.

 

European regulators sidelined on Anthropic superhacking model

European regulators have expressed concern over being "sidelined" regarding Anthropic’s unreleased AI model, Claude Mythos, which possesses advanced "super-hacking" capabilities.

While the model has been shared with a select group of 12 cybersecurity firms and 40 other organizations for defensive stress-testing, many European oversight bodies have not been granted direct access.

 

Key Tensions with European Regulators

Lack of Direct Access: Germany’s national cybersecurity agency, BSI, and other EU cyber officials have noted they have not yet directly tested the tool, receiving only "meaningful insight" through dialogues with developers.

Jurisdictional Limits: Because the model has not been officially "placed on the market" in the EU, it does not yet trigger many of the binding rules under the EU AI Act.

Security Implications: Claudia Plattner, head of the BSI, warned that the model’s power has "profound implications for national and European security and sovereignty".

Concerns Over Precedent: Experts like Laura Caroli worry that this sets a precedent where European officials are "at the mercy" of private U.S. tech firms for security oversight.

 

Regulatory Response & Endorsements

Staged Rollout Endorsed: Despite the lack of direct oversight, the European Commission has publicly welcomed Anthropic’s decision to delay the general release of Mythos, given its potential for large-scale cyber risk.

Active Dialogue: The EU's AI Office is reportedly in contact with Anthropic under the EU's code of practice to ensure future compliance with European standards once the model eventually hits the market.