OpenAI Publishes Guidance on Responding to Next-Frontier Cyber Capabilities

OpenAI has released a detailed policy piece outlining how it intends to respond when its models reach thresholds of critical cyber capability — effectively codifying a new category of AI risk evaluation. The document describes internal processes for identifying when a model crosses into territory where it could provide meaningful uplift to malicious cyber actors. This is directly tied to the decision to pause its latest model, making it both a policy statement and a real-world case study. For developers building security tooling or working in regulated environments, this framework is a useful reference for understanding how frontier labs are operationalizing responsible deployment. The guidance also signals that cyber capability benchmarks may become a standard part of AI model evaluation across the industry.