Skip to content
InfoNewsLLM-specific

Managing frontier AI risk: From training to deployment

Published
Record updated
View JSON

Summary

OpenAI describes how it manages frontier AI risk across the model lifecycle, from training through deployment, using defense in depth with multiple layers of protection. It says it evaluates both models and safeguards at each stage, requiring a safety review before advancing, and that it relies more on sandboxing and monitoring during research and more on alignment and system-level controls as models gain broader access.