Junious Media NetworkMonday, November 16, 2026
Juni/Tech
Advertisement
Security

OpenAI launches Daybreak Red Team to find and patch application flaws

OpenAI's Daybreak tools use AI agents to test web applications for security flaws and prepare patches for engineers to review before code is merged.

Key points

  1. OpenAI's Daybreak Red Team deploys AI agents to test web applications, validate vulnerabilities, and confirm whether patches stop exploits.
  2. Defense Factory takes confirmed findings and prepares code patches that engineers review and test before they are merged.
  3. OpenAI has pledged $1 billion toward subsidized Daybreak access, training, partnerships, and technical support, targeting utilities, governments, nonprofits, and open-source projects.
  4. Pricing for both tools and a general release date for Defense Factory have not been stated on OpenAI's official pages.

OpenAI has introduced two agent-based security tools under its Daybreak cybersecurity program. The first, Daybreak Red Team, uses a coordinated group of agents to probe web applications for security flaws. The second, Daybreak Defense Factory, takes those findings and prepares code patches for engineers to review and test before anything is merged. Both tools are available to verified, approved users. Defense Factory is currently in research preview.

The release adds specific products to what OpenAI calls its cybersecurity program. The program brings together AI capabilities, security solutions, vetted partners, and support for defenders. The company said applicants must be reviewed and approved before they can begin. The most advanced features require a separate application process.

What OpenAI's Daybreak Red Team does

Daybreak Red Team is built for penetration testing, the practice of probing software the way an attacker would to find weaknesses first. According to OpenAI's Daybreak program page, a coordinated group of agents tests web applications, confirms that each reported vulnerability is real, and checks whether a proposed fix actually stops the exploit from working.

The agents use source code and application context to map out possible attack routes through a system. Teams can point them at findings that already exist, feeding in results from code scanners, bug bounty reports, or coordinated disclosure submissions. That approach lets Red Team act as a validation layer rather than starting every assessment from scratch.

Scope control is a core part of the product. OpenAI said teams can specify which applications and endpoints agents are allowed to reach. Isolated environments and a Guardian review process enforce those limits. Without clear boundaries, an agent tracing an attack path could inadvertently reach systems outside the authorized test.

Advertisement

Red Team sits within Daybreak Red, the more restrictive of two access tracks. OpenAI said Daybreak Red covers authorized offensive work: red-team exercises, penetration tests, exploit validation, and controlled research conducted under strict oversight. It carries separate approval requirements beyond standard Daybreak access.

Advertisement

How Defense Factory moves a finding toward a fix

Where Red Team locates problems, Defense Factory is built to help resolve them. OpenAI said the product's agents trace each vulnerability through the codebase and develop a patch. Engineers then review and test that patch before the code is merged. Teams can bring in findings from their own security tools or route results directly from Red Team.

OpenAI framed the goal plainly on its Defense Factory page: "Turn your security backlog into fixes," the company said. Security teams often carry a long list of known issues that go unresolved. The usual reason is a shortage of engineering capacity to investigate and repair each one. Defense Factory is built to narrow that gap by handling the initial investigation and drafting a fix automatically.

For each finding that looks credible, Codex Security can reproduce the issue where safe and practical. It can write a focused patch and show evidence the change addresses the original problem. The workflow may also include a regression test that fails before the patch is applied and passes afterward, giving engineers a clear way to confirm the fix holds.

Advertisement

Human oversight stays at the center of the process. As OpenAI noted in a developers blog post, "The decision to apply the change still belongs to an engineer." Each proposed patch requires human inspection and approval before it touches the codebase. Defense Factory is in research preview, and OpenAI said a scan of a larger repository may take several hours to complete.

The wider Daybreak program and what it covers

Red Team and Defense Factory are two parts of the larger Daybreak program. OpenAI divides the program into two main tracks. Daybreak Blue covers defensive work: vulnerability discovery, validation, remediation, threat modeling, and security investigation. Daybreak Red is the authorized offensive track described above.

A joint effort called Patch the Planet brings together OpenAI, Trail of Bits, Calif, and HackerOne. Its aim is to help open-source maintainers validate findings, test patches, and coordinate disclosure. Many open-source projects lack a dedicated security team, so outside support can fill a real gap.

OpenAI has pledged $1 billion toward subsidized access, training, partnerships, and technical support through the Daybreak program, the company said on its Daybreak page. Utilities, state or local governments, community banks, nonprofits, and open-source projects are among the groups invited to register interest in that support.

Advertisement

The underlying technology is already in wide use. Codex Security Cloud had, in previously reported results, reviewed more than 30 million commits spanning over 30,000 codebases, OpenAI said in a developers blog post. Those figures show the system had handled broad, real-world use before Red Team and Defense Factory became available to outside organizations.

Advertisement

Who can apply and how approval works

Access to Daybreak requires verification. OpenAI said applicants must be reviewed before approval is granted. Advanced capabilities are reserved for users doing authorized work. The company matches safeguards to the activity, with Daybreak Red requiring a separate approval step beyond the standard process.

Organizations interested in either tool can apply through the Daybreak program page. For Defense Factory, the product page describes how to connect a repository and bring in existing security findings once access is in place. Users focused on defensive work fall under Daybreak Blue. Those running authorized offensive tests must qualify under Daybreak Red.

The approval model is meant to prevent misuse. Red Team agents follow attack paths through live applications. OpenAI said it limits those capabilities to users working within an authorized and clearly defined scope, not running unauthorized tests against systems they do not own.

What is still not known

Several practical details remain open. OpenAI has not stated pricing for Daybreak Red Team or Defense Factory on the official pages it has published. It is also not known how many organizations have been approved for access, or when Defense Factory will move out of research preview.

For security teams weighing whether to apply, cost and timeline are the most immediate unknowns. Research preview status means the product is still being developed. Its features and limits may change before it reaches a wider release.

Organizations that OpenAI specifically invited, including community banks and nonprofits, do not yet have a clear picture of what participation would cost them. How the $1 billion subsidy is structured, who qualifies for the most support, and what the approval timeline looks like in practice are questions the published pages leave open.

Advertisement

Sources