ai regulation / eu / anthropic

EU officials plan to meet with Anthropic to seek access to its Mythos model after the company warned the model is highly capable of finding cybersecurity vulnerabilities and could pose significant risks. The move reflects the EU’s interest in assessing potential threats and oversight needs; sources say any final access decision may require US government approval. This outreach signals regulators’ desire for direct technical review of advanced AI systems with dual-use capabilities, and highlights cross-border governance and national security complexities as more powerful models emerge.

The EU Commission is reportedly planning talks with Anthropic to gain access to its Mythos AI model after Anthropic warned Mythos is highly capable at finding cybersecurity vulnerabilities and could pose significant risks. Bloomberg says Brussels has sought access to assess the model, but release may require U.S. approval. Anthropic plans broader releases of new models in coming weeks with cybersecurity capabilities comparable to Mythos. The negotiations contrast with OpenAI, which has already granted the EU access to its GPT-5.5-Cyber model. The outcome matters for EU oversight of powerful cybersecurity-capable models and cross-border controls on sensitive AI tools.