OpenAI pauses GPT-6.1 Astra releases pending safety standards

openai.com11 min read

OpenAI details its framework for evaluating and managing frontier AI risks across the research, internal deployment, and public deployment stages. The company explains how it uses a defense-in-depth approach combining alignment, monitorability, security sandboxing, and automated controls to restrict model access and detect misalignment. These processes include pausing model training or releases when safety standards are not met. Open article →