OpenAI described Cogent Security’s agent workflow for investigating and addressing security issues. A context graph links assets, identities and dependencies, while TRACE’s deterministic tool policies and gateway authorization restrict execution. Higher-risk work uses sandboxing and customer approvals. The customer story cites a 75% reduction in mean time to remediate critical issues at one automotive enterprise and fivefold triage throughput at another customer. These are vendor-presented customer results, not a general performance guarantee. The operational lesson is the combination of contextual investigation with explicit execution boundaries.