OpenAI wants its models on the attacking side of corporate security, with permission. At its Cyber at the Frontier event on Thursday, the company introduced Daybreak Red Team, a product that puts a coordinated team of AI agents to work testing a company's web applications, and Daybreak Defense Factory, which turns the resulting findings into patches.

"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 President Greg Brockman said in his keynote, according to The Register.

What is OpenAI Daybreak Red Team?

Daybreak Red Team is an OpenAI product that sends a coordinated team of AI agents to penetration-test a company’s web applications, confirm vulnerabilities and check whether patches stop an exploit. It is available in early access by request.

OpenAI's product page says the agents use source code and app context to investigate attack paths and confirm vulnerabilities. Red Team can also test findings that arrive from code scanners, bug bounty submissions and disclosure reports, then rerun the attack after a fix to see whether it holds. Assessments run in isolated environments, and a review layer OpenAI calls Guardian limits which apps and endpoints the agents can reach.

Defense Factory handles the repair side. Its agents trace a vulnerability through the codebase and develop a patch, which engineers review and test before merge. SecurityWeek reported that it can import findings from Wiz, Snyk and Endor Labs, scan connected GitHub repositories, watch new commits and run scheduled assessments. Engineers can adjust a fix with Codex and open a draft pull request. Patches are not deployed automatically.

The Register reported that the agents run on GPT-6 Cyber, OpenAI's current cyber model. Access is gated: OpenAI says on its Daybreak program page that it verifies applicants and reviews their intended use before approving them, and that it has committed $1 billion in subsidized Daybreak access, training and support for groups such as utilities, local governments, community banks, nonprofits and open-source projects.

The launch landed on a busy day for AI policy. The same Thursday, The Register noted, a coalition of nonprofits filed suit in San Francisco Superior Court seeking to stop OpenAI, Anthropic, Google, Meta and SpaceX's AI businesses from deploying systems more capable than their current best models. Our report on OpenAI's research control room covers how the company says it monitors its most capable models.

Key takeaways

What to watch next

Who gets through the Red Team early-access queue, whether OpenAI publishes pricing for either product, and how quickly security vendors such as Wiz and Snyk build deeper links into Defense Factory.