OpenAI has launched two new cybersecurity products that use its AI models to find weaknesses in companies' software and help fix them, as the company pushes further into the market for AI-powered defence.

The tools, Daybreak Red Team and Daybreak Defense Factory, were unveiled on Thursday, 12 November, at the company's Cyber at the Frontier keynote. Both sit within OpenAI's Daybreak security platform and are built on its latest cyber models, currently GPT-6 Cyber.

Daybreak Red Team: automated penetration testing

Daybreak Red Team is a penetration testing service that uses fleets of AI agents to probe a company's web applications for vulnerabilities. One agent coordinates the assessment, others investigate source code and application context to work out whether flaws can actually be exploited, and reviewer agents check the results so security teams can prioritise what to fix first.

Nancy Wang, who leads cyber-applied AI at OpenAI, described the service as delivering penetration testing at the scale of automated continuous scanning. Red Team is launching in early access with selected customers, and a waitlist is open.

Daybreak Defense Factory: closing the loop

Daybreak Defense Factory is aimed at the other side of the problem: clearing the backlog of known vulnerabilities. Its agents scan codebases, find flaws, validate them and prepare patches for engineers to review, working through the Daybreak plugin in Codex, OpenAI's coding agent.

Users can work with Codex to add context, review evidence, adjust patches and test changes, with a dashboard organised by severity, detailed vulnerability views and a scan history. OpenAI said the tools build on Codex Security, which already helps firms find and fix code vulnerabilities as code changes.

A warning attached to the launch

OpenAI president Greg Brockman used the keynote to warn organisations that their security now depends on the frontier model they use for defence and how well they use it. The message was blunt: companies that defend themselves with weaker models risk falling behind attackers who are already using the strongest ones.

OpenAI did not disclose pricing for either product. The launch comes as the company argues there is a limited window to put these capabilities in defenders' hands before they spread more widely, a point its own researchers have made in recent publications about the dual-use nature of cyber AI.

The announcement landed on the same day that four nonprofits sued OpenAI and other leading AI companies in San Francisco, asking a court to freeze development of more capable AI systems. OpenAI has not commented publicly on that lawsuit.

Sources: The Register and The Deep View.