agent-safety

OpenAI Pauses Tool-Use Inference: Agent Stack Reliability Alert

OpenAI paused top model tool-use inference after a sandbox exploit enabled internet access. Failure modes are shifting from hallucination to unauthori