Anthropic to offer EU access to its advanced Mythos model

Anthropic has announced that it will provide the European Union access to its advanced AI model, Mythos, following months of discussions driven by cybersecurity apprehensions. The European Commission confirmed on Monday that it had engaged in several productive dialogues with the American AI company. EU tech sovereignty spokesperson Thomas Regnier expressed optimism in a statement, noting, "We welcome the latest developments regarding potential future access," while emphasizing the need to better understand the risks associated with this powerful technology. Initially released to a select group of companies in April as part of Anthropic's cybersecurity initiative known as Project Glasswing, Mythos has demonstrated remarkable capabilities in identifying vulnerabilities and weaknesses within software systems. This rollout raised alarms among governments and tech firms regarding the potential for misuse by malicious actors. Regnier cautioned, "Let's not forget that Mythos is just one in a series of potent models set to enter the market. This presents a shared challenge, and we are deepening our collaborations with like-minded partners, including the United States." While the specifics of the access agreement remain unclear, discussions between the EU and U.S. administration have intensified in recent weeks. Sources indicated that Anthropic had informed the Commission that government approval was necessary for the model's deployment, with the U.S. government generally opposed to sharing such advanced technology with non-U.S. entities to maintain its leadership in the AI sector. In May, the EU had already gained access to OpenAI's GPT-5.5-Cyber model, but the Commission noted that its negotiations with Anthropic were at a different stage, involving four or five meetings. The release of Mythos in April had generated significant concern among various sectors due to its ability to expose numerous previously undiscovered software vulnerabilities, which could potentially fuel cybercrime if exploited by the wrong hands. Additionally, the U.S. government announced agreements in May with Google DeepMind, Microsoft, and Elon Musk's xAI, allowing for assessments of AI models prior to their public release.