OpenAI offers EU access to GPT-5.5 cyber variant after Anthropic locks out European officials
Tags AI · Security · Policy
OpenAI announced on May 11 that it is in talks with the European Commission to grant EU authorities access to a cyber-focused variant of GPT-5.5 capable of identifying software vulnerabilities. Former UK Chancellor George Osborne, leading the initiative, wrote to the Commission offering access to the most cyber-permissive version of the model, with main safeguards built in to prevent misuse as a hacking tool. The move contrasts with rival Anthropic, which has not granted EU access to its cyber-capable model Mythos, limiting it to a dozen US-based trusted tech firms and 40 unnamed organizations. A group of 30 lawmakers pressed the EU's cybersecurity agency ENISA to gain access to Mythos, and four countries including Spain requested more coordination. The European Commission welcomed OpenAI's transparency, saying it would help monitor the model and address security concerns.
Technical significance
OpenAI's offer to the EU creates a de facto regulatory advantage over Anthropic in the European market, where access to frontier cyber-capable models is becoming a prerequisite for national security readiness. The divergence in access policies between OpenAI and Anthropic sets up a competitive dynamic where governments may favor AI vendors that offer transparency, potentially reshaping the global AI governance landscape. The fact that a cyber-permissive model variant exists separately from the safeguarded main model also reveals how frontier AI companies are architecting dual-use capabilities.