ENISA will join Project Glasswing, Anthropic's program for testing the powerful vulnerability-finding model before a wider release
Anthropic CEO Dario Amodei (Krisztian Bocsi/Bloomberg via Getty Images)
Anthropic is set to give the European Union's cybersecurity agency, ENISA, access to Mythos, a powerful AI model designed to identify security flaws in software, according to Bloomberg. ENISA would become the first E.U. body to gain access to the model.
The AI company communicated the decision to the European Commission over the weekend, according to Bloomberg, citing unnamed sources. Through Project Glasswing, ENISA will be able to evaluate Mythos alongside a limited set of organizations before the model reaches a broader audience. An Anthropic representative declined to comment.
The European Commission confirmed it had held discussions with the company. "I can confirm that the Commission had several productive meetings with Anthropic. We welcome the latest developments on potential future access," Commission spokesman Thomas Regnier said, according to Reuters. The exact terms and timing of ENISA's access have not been made public.
Anthropic first previewed Mythos in April as part of Project Glasswing, according to CNBC. Its release set off a wave of alarm across the public and private sectors — from finance ministries to major technology companies — over the prospect that a tool capable of surfacing vast numbers of undiscovered software flaws could end up enabling the very attacks it was designed to defend against. Anthropic has described those risks as the reason for its measured rollout. So far, only a handful of recipients have been brought into the program: federal agencies along with some of the largest American banks and technology companies. Access was extended to the U.K.'s AI Security Institute not long after Mythos was first unveiled, according to Bloomberg.
Senior Commission officials made the trip to San Francisco last week to press Anthropic leadership for access in person, as European governments and finance ministries had grown increasingly impatient in the weeks since the model's April debut, according to Bloomberg. Bloomberg also reported that Washington had blocked at least some of Anthropic's earlier attempts to widen the program, on national security grounds. A complicating factor throughout the negotiations was Anthropic's position that clearance from Washington was a prerequisite to any transfer of access to foreign governments, CNBC had previously reported.
Regnier framed the development as part of a broader challenge facing regulators. "Mythos is not one off, a new wave of powerful models are coming to the market," he said. "This is a shared challenge, and we are intensifying our discussions with like-minded partners, including the United States."
The E.U. had previously been granted access to OpenAI's GPT-5.5-Cyber model in May, according to CNBC.
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
