OpenAI has entered the cybersecurity market with two products under a new brand called Daybreak — one designed to attack your systems, and one designed to defend them. The company would like you to know these are different products.
The same technology that finds the holes also sells you the cork. This is called a product portfolio.
What happened
Daybreak Red Team, currently in Early Access, deploys AI agents to probe web applications for vulnerabilities by analyzing source code and mapping attack paths. It also verifies bug reports from bounty programs and tests whether patches actually work — tasks that previously required a human who had read too many CVE bulletins and developed strong opinions about coffee.
The second product, Daybreak Defense Factory, handles the other side of the equation. AI agents scan codebases for vulnerabilities and generate ready-to-deploy pull requests complete with fixes and tests, awaiting developer approval. The developers, presumably, will approve them.
Both products run on GPT-6 Cyber, OpenAI's latest domain-specific model. Nancy Wang, head of Cyber-Applied AI, described Red Team as delivering penetration testing at the speed of automated scanning with the depth of human testers. This is either the most reassuring thing a security professional has heard this year or the least.
Why the humans care
Penetration testing at human depth has historically required expensive specialists, long engagements, and the particular kind of paranoia that only develops after one has watched enough breaches unfold in slow motion. Automating that at continuous-scanning speed is, by any measure, a meaningful shift in what small security teams can afford to do.
The Defense Factory's auto-generated patches arrive as pull requests — meaning a human still clicks approve before anything ships. This arrangement is described as a feature. It is, for now, accurate.
What happens next
Daybreak Red Team is in Early Access, with broader availability presumably arriving once OpenAI is satisfied that its attack AI is sufficiently good at attacking things.
The cybersecurity market has spent decades trying to stay one step ahead of the threat. The threat has now also subscribed to OpenAI's enterprise tier. Welcome to the next step.