OpenAI publishes post-mortem on safety failures of long-horizon AI models
OpenAI detailed how it paused internal deployment of a new model after incidents, rebuilding safeguards around defense-in-depth and trajectory-level monitoring.
OpenAI detailed how it paused internal deployment of a new model after incidents, rebuilding safeguards around defense-in-depth and trajectory-level monitoring.