OpenAI ships Daybreak Red Team and Defense Factory agents

OpenAI opened a Thursday keynote by Greg Brockman with two product announcements and a plain sales pitch: companies that don't put a frontier model on defense aren't going to be secure. "We're now at a point where companies will be secure only to the extent of the frontier model that they use to defend themselves and how well they utilize that," OpenAI's president said, according to The Register.

The first product, Daybreak Red Team, is in early access. It deploys agent fleets that ingest source code and application context to pen-test web apps, and will also validate flaws reported by scanners and bug bounty researchers and retest patches. Nancy Wang, who leads cyber-applied AI at OpenAI, pitched it to the room as a "superhuman pen-test at the scale of automated continuous scanning."

The second, Daybreak Defense Factory, runs on agents built on OpenAI's latest cyber model, which The Register identifies as GPT-6 Cyber. It scans codebases and prepares patches as merge-ready pull requests, with engineers reviewing proposed fixes before deployment. OpenAI frames the loop as "finding, validating, and fixing vulnerabilities at scale."

The write-up names no design-partner customers, publishes no benchmark numbers, and does not say where customer source code sits during a Red Team run.

It lands a day after our coverage of a different autonomous pen-test tool, AntiFox, that pulled a live AWS production secret during a run, and in a 90-day stretch that has produced 411 tracked OpenAI stories on our feed.