Anthropic has granted the European Union Agency for Cybersecurity (ENISA) access to its Mythos AI safety evaluation platform, the company announced today. The move comes through Project Glasswing, a collaboration described as the result of "strong bilateral cooperation" between the European Commission and the AI developer.
ENISA, the EU's primary cybersecurity agency, will use Mythos AI to assess potential risks posed by artificial intelligence systems. The tool is designed to probe large language models for vulnerabilities, bias, and other failure modes before they are deployed at scale.
Details on how ENISA will operationalize the platform remain limited. Anthropic has previously used Mythos AI internally to stress-test its own models, but this marks the first time the tool has been shared with a regulatory body for independent evaluation.
The partnership reflects a broader push by Brussels to integrate AI safety testing into its cybersecurity framework. Project Glasswing itself is a relatively new initiative focused on closing gaps between AI developers and European regulators.
No specific timeline has been provided for when ENISA will begin using the tool or publish its findings. The announcement comes amid escalating global debate over how—and by whom—advanced AI systems should be audited.