According to Bloomberg, European Commission officials traveled to San Francisco on May 28 to hold talks with Anthropic, seeking more information about the capabilities of the Mythos model and pushing to open access to the model for the European Union. Sources indicated that the talks are still in an information-gathering phase, and no final arrangements have been made. Mythos was released in April this year and is capable of autonomously discovering and exploiting software zero-day vulnerabilities. Anthropic currently only grants preview access to a small number of U.S. partners, leaving European financial institutions and critical infrastructure operators unable to access it.
The talks represent the latest step in the EU’s ongoing pressure campaign. In early May, EU Economy Commissioner Valdis Dombrovskis confirmed that the EU has been in contact with Anthropic; in late May, Spain’s Economy Minister warned that negotiations have made “limited progress.” Meanwhile, OpenAI has proactively committed to granting the European Commission access to its cybersecurity model GPT-5.5-Cyber. EU Commission spokesman Thomas Regnier also stated that once the AI Office’s enforcement powers take effect in August this year, it will “secure access to Mythos” if necessary. Anthropic announced on the same day that the Mythos model is expected to be made available to all customers “within the coming weeks.”