European regulators have largely been left out of early access to Anthropic’s Mythos model, a cybersecurity-focused AI system that the company says can identify and exploit technical vulnerabilities beyond the ability of most humans, according to a Politico report.
KEY FACTS
- Access limited Anthropic has initially restricted Mythos to a small group of selected companies.
- Named participants Apple, Microsoft and Amazon are among the firms given access to analyze the model.
- UK testing The UK AI Security Institute has tested the system and acted on its results.
- EU gap Germany is reported to have started talks with Anthropic, but it has not yet gained access.
Anthropic has framed the rollout as part of Project Glasswing, a limited-access program meant to reduce security risks before wider release. The company has not opened the model broadly, and the report says that leaves public authorities with limited influence over how the technology is assessed and shared.
Mythos is described as a system aimed at cybersecurity use cases. The disclosure says it can find and exploit technical weaknesses at a level that exceeds most human testers, which raises questions about how quickly such tools should be distributed and who should review them first.
Germany’s chief cybersecurity official, Claudia Plattner, told Politico the question of whether a tool of such extraordinary power will reach the open market is pressing. She said that issue has major implications for national and European security and sovereignty.
WHY IT MATTERS
The limited access highlights a broader tension in advanced AI security work, where private companies may control early testing of powerful systems before regulators do. That can affect how quickly governments identify risks and shape rules for tools that could be used to find or exploit vulnerabilities.