OpenAI is in active talks with the European Commission to grant access to its most advanced cybersecurity-focused AI model, one capable of identifying software vulnerabilities. The move positions the ChatGPT maker as the first major AI lab to open its cyber capabilities to EU regulators, who have been struggling for weeks to assess the security risks posed by frontier AI systems.
The timing is pointed. Anthropic, OpenAI’s chief rival in the safety-conscious AI race, has not yet authorized the EU to access its own cybersecurity model, Mythos.
What the model actually does
OpenAI’s cybersecurity model, referred to as GPT-5.5-Cyber, is specifically engineered to identify software flaws and simulate intrusions.
As of May 1, GPT-5.5 completed a full simulated corporate network hack, making it only the second AI system to accomplish that feat. The first was Anthropic’s Mythos. The two models now appear to be roughly matched in their ability to break into simulated enterprise environments.
Under the AI Act, European regulators need to evaluate the cybersecurity risks that advanced AI models introduce. By granting direct access, OpenAI is effectively letting the Commission kick the tires on its most capable cyber tool.
Why crypto should be paying attention
Financial losses from crypto hacks exceeded $1.5 billion in 2025. If AI models can simulate corporate network hacks, they can also be pointed at smart contracts, bridge protocols, and DeFi platforms.
Recent months have already seen reports of scams exploiting AI agents for crypto theft. The threat landscape is evolving in real time, with attackers leveraging AI tools to automate phishing, social engineering, and exploit discovery.
AI-related tokens saw a 5% price increase following the announcement of these cybersecurity AI advancements.
The regulatory chess match
OpenAI’s overture to the EU is as much about strategy as it is about security. The AI Act represents the most comprehensive regulatory framework for artificial intelligence anywhere in the world. By volunteering access to its most capable cyber model, OpenAI is making a calculated bet: the company gets to demonstrate transparency and build goodwill with regulators, while gaining influence over how the EU ultimately classifies and regulates AI systems with offensive cyber capabilities.
For Anthropic, if the EU ends up writing its cybersecurity AI guidelines based primarily on its experience with OpenAI’s model, Anthropic risks being evaluated against a framework it had no hand in shaping.
The $1.5 billion lost to crypto hacks in 2025 is the baseline. The DeFi sector in particular should be watching closely, as projects such as Fetch.ai have already begun integrating AI technologies for automatic security audits.