Anthropic is opening the door for Europe’s cybersecurity agency to use what may be the most powerful vulnerability-hunting AI tool ever built. The company will onboard ENISA, the EU’s cybersecurity agency, into Project Glasswing, granting access to its Claude Mythos Preview model.
It’s the first time an EU agency has gained access to an AI system of this caliber before a broader rollout. And the timing is anything but coincidental.
What Mythos actually does
Claude Mythos Preview, announced in early April 2026, isn’t your standard large language model with a security plugin bolted on. The system can autonomously identify nearly 10,000 severe software vulnerabilities and execute complex attack simulations that would traditionally require teams of human security researchers working for months.
That capability is precisely why Anthropic has kept the model on an extremely short leash. Access has been limited to vetted US-based organizations and select international partners, including Amazon, Apple, Microsoft, and JPMorgan Chase.
And that concern isn’t theoretical. In April 2026, reports surfaced of unauthorized access claims tied to a third-party vendor, prompting a formal investigation by Anthropic.
How the EU deal came together
The negotiations between the European Commission and Anthropic have been substantial. As of May 11, 2026, the two sides had held four to five meetings to hammer out the terms of access. Those talks started shortly after Mythos was announced in April and have been running in parallel with a separate, competing offer.
OpenAI revealed its own cybersecurity-focused AI model offering on May 11, 2026, aimed at the same kind of institutional clients.
European firms and financial institutions have been vocal about their concerns over competitive disadvantages. If US companies and agencies get access to these tools months or years before their European counterparts, the security gap widens in ways that matter for everything from critical infrastructure to financial services.
Why the crypto market should be paying attention
Mythos’s capabilities extend well beyond finding bugs in traditional software stacks. The model can identify vulnerabilities in DeFi infrastructure that go far deeper than the smart contract exploits that have dominated headlines for the past few years, including protocol-level weaknesses, bridge vulnerabilities, and systemic risks in composable DeFi architectures.