OpenAI publishes its Defense Factory playbook
OpenAI says more than 250 people strengthened defences across hundreds of systems, using cyber models to find vulnerabilities, validate them and verify fixes. The important detail is the loop: discovery does not end at a finding. It carries through validation and proof that the remediation held. That is a useful description of where agents belong in security operations: inside a controlled path with an observable result, not as a detached answer engine.
Our takeThis is the right frame for persistent agents. Bounded authority, evidence at every stage and a verifiable finish are the operating contract.
