TechAILatest.newsTrackers
Cybersecurity / AI security tools

OpenAI launches Daybreak Red Team and Defense Factory agents to pen-test apps and draft fixes

OpenAI unveiled Daybreak Red Team, agent teams that pen-test web apps, and Daybreak Defense Factory, which drafts fixes, at its 12 November cyber event.

By Gürkan E.Published

Abstract illustration of a shield over a networkTech AI Latest
Illustration of a shield over a network, made by Tech AI Latest for this story. Illustration: Tech AI Latest (Original illustration)

OpenAI launched two security products on 12 November 2026 at its Cyber at the Frontier event. Daybreak Red Team, in early access, uses coordinated AI agents to pen-test web applications, confirm reported flaws and retest patches. Daybreak Defense Factory, a research preview plugin, takes findings from security tools or Red Team and prepares fixes for engineers to review. OpenAI published no prices and no accuracy or false-positive figures for either product.

Key facts

  • Announced: 12 November 2026 at OpenAI's Cyber at the Frontier keynote (OpenAI Daybreak)
  • Daybreak Red Team: early access by request; agent teams test web applications, confirm reported vulnerabilities and retest patches
  • Daybreak Defense Factory: research preview plugin that connects to a GitHub repository and prepares proposed fixes
  • Human step: Defense Factory prepares changes but does not deploy them; engineers review the fixes
  • Not stated: pricing, accuracy or false-positive rates, and the model behind each product

What OpenAI launched

OpenAI used its Cyber at the Frontier livestream on 12 November 2026 to add two products to Daybreak, the security program it started in May. Both move Daybreak from a model that helps a human researcher toward agents that run parts of the security workflow on their own, with people approving the results.

Daybreak Red Team is an offensive testing service for web applications. According to OpenAI's launch material as reported by SecurityWeek and Help Net Security, it runs a team of agents: one coordinates the assessment, others read the application's source code and context to look for attack paths and confirm that a vulnerability is real, and reviewer agents check the findings. OpenAI says the same agents can validate flaws reported by scanners or bug bounty researchers and retest a patch to see whether it actually blocks the exploit. Red Team is in early access. Organizations fill in a request form, and registering does not guarantee a place.

Daybreak Defense Factory sits on the repair side. It investigates findings from a company's existing security tools, or from Red Team, and prepares proposed fixes. OpenAI's developer guide to Daybreak tells teams whose repositories need ongoing attention to install Daybreak Defense Factory (Preview) from the plugin marketplace and connect the GitHub repository. Defense Factory prepares changes but does not deploy them; engineers review each proposed fix before it ships. It is a research preview, and access requires a workspace that already has Daybreak access.

How it fits the rest of Daybreak

Daybreak launched in May 2026 as OpenAI's answer to defensive security work, with partners such as Cisco, Cloudflare, CrowdStrike and Palo Alto Networks building it into their own products. It is split into two lanes: Daybreak Blue for common defensive tasks on OpenAI's mainline models, and Daybreak Red for approved organizations doing more sensitive work such as penetration testing and exploit validation. In September OpenAI added a $1 billion commitment of subsidized access, training and support for organizations that protect essential services.

The new products arrive a week after OpenAI released GPT-6 Cyber to vetted Daybreak Red customers, a model it rates Critical for cyber capability, as we reported on 6 November. Coverage of the keynote does not agree on which model powers Red Team and Defense Factory, and OpenAI's own pages, as quoted in that coverage, do not name one, so this story does not either. Defense Factory builds on the patching work OpenAI already offered through Codex and its Codex Security plugin.

What changes for security teams

Until now Daybreak gave approved teams access to models and tools that a human researcher drove. Red Team packages the full loop of a web application pen test, from finding a weakness to proving it and checking the fix, as a managed agent product. Defense Factory turns findings that sit in a scanner backlog into draft patches inside the repository where developers already work. Together they target the gap that critics of AI vulnerability discovery have pointed to since May: models can now find flaws faster than teams can confirm and patch them.

OpenAI is not alone in this market. Anthropic offers Claude Mythos 5.2 only to vetted organizations, and security vendors are building their own agent stacks; Cogent described its plan for a fully autonomous Cyber Defense Factory when it agreed to buy Specular on 10 November. OpenAI has also brought outside testers into its own systems, including the startup Hacktron, as covered here on 8 November. More cybersecurity news is on the hub page.

What to watch next

OpenAI has not published a price for either product, and it has given no accuracy or false-positive figures, so buyers in the early access program will have to measure how many confirmed findings and how many mergeable patches the agents produce. The other open questions are the scope controls on Red Team, since an agent that probes live web applications needs strict limits on which systems it may touch, and when Defense Factory leaves research preview. Independent results from early customers would be the first test of OpenAI's claims.

What changed

BeforeDaybreak gave approved teams OpenAI models and Codex Security tools that human researchers drove; no managed pen-testing or automated fix product
AfterDaybreak Red Team (early access) runs agent teams that test web apps and retest patches; Daybreak Defense Factory (research preview plugin) drafts fixes in connected GitHub repositories for engineers to review
Who it affectsApplication security teams, penetration testers, organizations already in Daybreak, and vendors of vulnerability scanning and remediation tools

Questions people ask

What is Daybreak Red Team?
An OpenAI early access product in which coordinated AI agents pen-test web applications, confirm reported vulnerabilities and retest patches. Organizations request access through OpenAI's Daybreak page.
Does Daybreak Defense Factory deploy fixes automatically?
No. It prepares proposed fixes in a connected GitHub repository, and engineers review them before anything is deployed. It is in research preview.
How much do the new Daybreak products cost?
OpenAI did not publish pricing for Daybreak Red Team or Defense Factory at launch.

Sources

  1. Daybreak (OpenAI, 12 Nov 2026)
  2. Scaling cyber defenders with Daybreak (OpenAI, 12 Nov 2026)
  3. OpenAI Unveils Daybreak Red Team, Defense Factory Security Tools (SecurityWeek, 12 Nov 2026)
  4. OpenAI's Daybreak brings AI agents to security testing and vulnerability fixes (Help Net Security, 13 Nov 2026)
  5. OpenAI launches Daybreak Red Team and Defense Factory to fix vulnerabilities using AI (Neowin, 12 Nov 2026)

Reference pages

Spotted an error? See the corrections policy and tell the editor. Facts in this story come from the sources listed above; vendor claims are attributed.

Comments

Loading comments.

No links, no insults; comments that break the rules are not published. 0/1000

Related stories

All cybersecurity