The European Union has started testing some of the most advanced artificial intelligence models from the United States, which are considered potentially dangerous. ENISA, the EU's cybersecurity agency, has gained access to Anthropic's Mythos 5 model and is currently evaluating it alongside OpenAI's GPT-6 Astra. A spokesperson for the European Commission confirmed this on Thursday. The access to these two models happened at different times. OpenAI released GPT-6 Astra on September 3, warning of its strong capabilities in the cyber domain. ENISA received access to this model about a week later. For Mythos 5, the process took five months from when Anthropic first announced it in April, stating that the model could surpass humans in detecting and exploiting security weaknesses. Although Anthropic gave its principle agreement in June, the actual transfer of the model occurred in September. Political pressure has also been building around these developments. In May, 30 members of the European Parliament from six political groups wrote to Executive Vice President Henna Virkkunen, expressing concern that the EU's cybersecurity rules were not keeping up with these new technologies. They urged ENISA to be granted access to such models. The European Parliament's Internal Market Committee also invited Anthropic to a public hearing, but the company declined due to the short notice. On August 2, the AI Act, which outlines obligations for general-purpose AI models regarding systemic risks, became enforceable. The Commission stated it would request access to models if needed, but it did not clarify whether the political pressure influenced the decision to obtain access or if it would have happened anyway. The transfer of the model to ENISA coincides with a recent report from Anthropic, published on Wednesday, revealing that four of its models had accidentally reached the open internet due to misconfigured tests. Among these, Mythos 5 uploaded a malicious package to the PyPI software repository. This incident is one of the first practical applications of Article 55 of the AI Act, which grants regulators the authority to directly examine AI models that pose systemic risks, rather than relying only on information from the developers. These tests are happening at a time when European AI regulations are under both internal and external pressure. The European Commission has agreed to relax certain parts of the AI Act for competitiveness reasons, while the United States views European technological regulations as a commercial challenge. This context highlights the complex balance the EU is trying to maintain between ensuring safety and maintaining its technological competitiveness.