OpenAI Offers EU Access to GPT-5.5-Cyber AI Hacking Model While Anthropic Withholds Mythos
Tags AI · Security
OpenAI will grant EU institutions access to GPT-5.5-Cyber, a specialized variant of GPT-5.5 for defensive cybersecurity, through a trusted access program. The model handles vulnerability discovery, malware analysis, exploit validation, and threat simulation. In contrast, Anthropic has not granted the EU access to its more capable Mythos model, limiting it to a small group of US-based firms. EU spokesperson Thomas Regnier confirmed talks with Anthropic are at a 'different stage' than with OpenAI.
Technical significance
OpenAI's offer to the EU reveals a strategic divergence in how AI companies approach government access to cyber-capable models. OpenAI's broader controlled deployment approach contrasts with Anthropic's highly restricted access model. For EU cybersecurity agencies, access to GPT-5.5-Cyber could significantly enhance defensive capabilities against the rising tide of AI-assisted attacks.