OpenAI Negotiates EU Access to Cyber Model

According to CNBC, the European Commission is in discussions with OpenAI over preview access to the cyber-focused model GPT-5.5-Cyber. CNBC reports that OpenAI announced GPT-5.5-Cyber is rolling out in limited preview to vetted cybersecurity teams. European Commission spokesperson Thomas Regnier was quoted by CNBC saying, "We welcome OpenAI's transparency and intent to give commission access to new model," and confirming that an exchange had taken place and that further discussions were planned. CNBC also reports that Anthropic released the model Mythos about a month ago but has not granted the Commission preview access; Regnier said discussions with Anthropic are at a "different stage." CNBC adds that both companies were approached for comment.

Governments seeking preview access to high-capability models increasingly require samples or controlled previews to evaluate misuse risk and defenses. Observers of recent EU engagement with AI firms note regulators often prioritise models that have explicit cyber-capability features for closer scrutiny because of potential to generate exploit code or operational guidance.

For practitioners, regulated-preview workflows can change how security teams and vendors coordinate. Companies deploying or evaluating models with offensive cyber capabilities commonly adopt staged rollouts, vetting programs, and technical guardrails to allow third-party review while limiting public exposure. These patterns affect incident-response playbooks, red-team exercises, and requirements for reproducible safety tests.

Track whether the Commission obtains hands-on preview access to GPT-5.5-Cyber and whether Anthropic grants similar access for Mythos; those outcomes will indicate the EU's operative approach to high-risk capabilities. Also watch for any technical or policy findings the Commission publishes from reviews, and for changes in vendor preview processes that follow regulatory feedback.