Timeline

OpenAI launches Daybreak security products for red-teaming and defence

Red Team, in early access, uses teams of agents to test web applications; Defense Factory has agents write patches that engineers review before merge. The Register reported both run on GPT-6 Cyber.

  • Security & misuse
  • Labs & people
  • Minor

OpenAI launched two products under its Daybreak security platform: Daybreak Red Team and Daybreak Defense Factory. The Register reported the launch from OpenAI president Greg Brockman’s Thursday keynote on 12 November. OpenAI’s own product pages for the two are undated.

Red Team, offered in early access, is for penetration testing of web applications. OpenAI’s page says a coordinated team of agents uses source code and application context to investigate attack paths and confirm vulnerabilities. It can also test findings from code scanners, bug-bounty submissions and disclosure reports, and check whether a patch stops an exploit. OpenAI says scope is enforced with isolated environments and “Guardian review”, with controls over which applications and endpoints agents can reach.

Defense Factory takes findings from a customer’s existing security tools, or from Red Team, and has agents trace each vulnerability through the code and prepare a patch. OpenAI’s page says engineers review and test proposed patches before merge. The Register reported that the agents are built on GPT-6 Cyber, the model OpenAI released to vetted defenders a week earlier; the product pages do not name a model. Nancy Wang, who leads cyber-applied AI at OpenAI, was quoted by the Register as saying engineers get a “merge-ready pull request with the patch and a test that shows that it works”.

Brockman told the audience, in the Register’s account, that “companies will be secure only to the extent of the frontier model that they use to defend themselves”, while allowing that security fundamentals still matter. That is a vendor’s pitch, and the Register’s piece took a sceptical tone towards it. It noted that the keynote coincided with a lawsuit by a coalition of nonprofits seeking to stop OpenAI and four other developers from building more capable models, which cites incidents involving OpenAI agents.

No pricing, customer figures or measured results for either product were given in the sources reviewed. Daybreak began in May as a programme pairing OpenAI’s models with partner organisations to find and patch vulnerabilities; these two products package that approach for individual customers.

In the commentary

What people were saying around this time — external links, from the record's commentary rail.