What changed
OpenAI launched Daybreak Red Team in early access and Defense Factory in research preview. Red Team coordinates agents to test authorized web applications, validate findings and retest patches; Defense Factory connects to GitHub to investigate vulnerabilities, prepare reviewable fixes and monitor commits or schedules.
Why it matters
The release turns a frontier cyber model into repeatable offensive and defensive workflows with isolated environments and review gates. That can shorten validation and remediation loops, but it concentrates repository, target and exploit authority in one agent system.
What to do
Start with an authorized disposable target and a read-only repository mirror, establish human-triaged baselines for false positives and missed findings, then require patch review and retesting before any merge or deployment.
Caveat
Red Team is early access and Defense Factory is a research preview. OpenAI has not published pricing or independent accuracy, false-positive or remediation-outcome evidence; scope authorization and human review remain mandatory.