OpenAI Launches Daybreak Red Team and Defense Factory Security Tools

What happened

Daybreak Red Team is in early access, and organizations apply to join. Its agents combine source code and application context to explore attack paths, validate vulnerability reports submitted by third parties, and retest applications after patches are applied. Nancy Wang, who leads cyber-applied AI at OpenAI, described Red Team at the keynote as "superhuman pen-test at the scale of automated continuous scanning," as reported by The Register.

Daybreak Defense Factory ships as a preview plugin inside ChatGPT for workspaces that already have Daybreak access. Its agents take findings from existing security tools and Red Team, trace each vulnerability to its root cause through the codebase, and prepare merge-ready pull requests with tests; engineers review the changes before merging, and the platform does not deploy them automatically. The workflow supports findings from Wiz, Snyk and Endor Labs, alongside connected GitHub repositories.

Both products run on OpenAI's cybersecurity models, including GPT-6 Cyber, which reached select approved Daybreak Red customers on Nov 5. The Daybreak platform launched on June 22 around GPT-5.5-Cyber, the Codex Security plugin and related tooling. Livestream coverage reported that an unreleased cybersecurity model scored 77.0% on ExploitGym, but the benchmark chart did not appear on the official product pages and the figure is not officially confirmed.

Positioning

OpenAI President Greg Brockman said at the keynote that organizations will be "secure only to the extent of the frontier model" they use for defense. Daybreak is positioned as handing a security team's existing workflow to agents: scanning, validation, patching and retesting as one continuous loop, rather than a one-off scan report.

The two products differ from the Codex Security CLI OpenAI open-sourced on Nov 11: the CLI is a package teams install, while Defense Factory is a hosted workflow that wires scanners, issue trackers, isolated development environments and Daybreak models into one loop. OpenAI has published no pricing for either product and no date for general availability.

Key facts

1. Launch: Daybreak Red Team and Daybreak Defense Factory were introduced on Nov 12 at the "Cyber at the Frontier" keynote
2. Red Team: Agents run authorized penetration tests on web applications, exploring attack paths, validating vulnerability reports and retesting fixes; in early access, by application
3. Defense Factory: Agents scan codebases, trace vulnerabilities to their root cause and prepare merge-ready pull requests with tests; engineers review before merging, with no automatic deployment
4. Integrations: Supports findings from Wiz, Snyk and Endor Labs and connected GitHub repositories; Defense Factory ships as a preview plugin inside ChatGPT
5. Models: Underlying models include GPT-6 Cyber, which reached select approved customers on Nov 5; the Daybreak platform launched on June 22
6. Pricing: Neither product has public pricing or a general-availability date
7. Benchmark: Livestream coverage reported an unreleased cybersecurity model scored 77.0% on ExploitGym; the chart did not appear on official product pages and the figure is not officially confirmed