Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
Policy & SafetyInsight · On RadarThe Decoder→ original

OpenAI Moves Into Cybersecurity With Two Daybreak Products That Find Bugs and Write Patches

OpenAI is entering the cybersecurity market with two Daybreak products that find vulnerabilities in code and prepare fixes themselves. Red Team, in early access, works like an automated penetration tester: it reads source code and maps possible attack paths against a web application. Defense Factory finds vulnerabilities in a repository and submits a ready pull request with a patch and tests. Both run on GPT-6 Cyber, and a developer makes the final call on any change.

AI-processed from The Decoder; edited by Hamidun News
OpenAI Moves Into Cybersecurity With Two Daybreak Products That Find Bugs and Write Patches
Source: The Decoder. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Decoder ↗
Editor-in-Chief
Key metric #1Red Team checks bug bounty reports and code scanner findings
Red Team's goalthe speed of continuous scanning and the depth of live penetration testers
Defense Factory merges nothing on its ownthe pull request waits for human approval

Most teams are bottlenecked by a shortage of people. Scanners spit out hundreds of warnings, penetration tests are commissioned once a year, and patches sit in the backlog for weeks. OpenAI has decided to target this gap: the company has launched two products under the Daybreak brand. In them, AI agents (programs that carry out a chain of actions on their own) look for vulnerabilities in code and prepare fixes. Both run on the new GPT-6 Cyber model, tuned for security tasks. The first product, Red Team, imitates the actions of an attacker. The second, Defense Factory, handles defense: it finds a hole and proposes a ready-made patch.

Daybreak Red Team: An Automated Penetration Tester

Red Team is currently available in early access. The task will be familiar to anyone who has commissioned a penetration test (pentest): working out how an attacker could break into a web application. The agents analyze the source code and the context of the application itself, then build a map of possible attack paths.

This is easiest to explain with a building. A conventional scanner checks each door against a list and reports: "the lock is weak." A live pentester looks at the whole building and reasons: you can't get through the main entrance, but you can through the service door and the forgotten window on the second floor. Red Team, judging by the description, tries to work in the second mode, only without breaks and weekends.

The tool has a second function that will be useful in practice. It checks vulnerabilities from bug bounty program reports (where outside researchers receive rewards for discovered holes) and findings from code scanners, and it also tests patches. Anyone who has triaged incoming reports knows that most of the time goes to the question "is this actually exploitable, or a false alarm?" If the agent takes over this routine, the team gets problems that are already confirmed.

Nancy Wang, head of the Cyber-Applied AI division, said the tool is designed to deliver pentesting at the speed of automated continuous scanning while matching the depth of human testers.

Daybreak Defense Factory: From Finding to Pull Request

Defense Factory covers the defensive side. AI agents scan the company's codebase, look for vulnerabilities, and write the fixes themselves. The output for the developer is a ready pull request: a code change along with accompanying tests. All that remains is to review the changes and approve or reject them, and there is no need to read a PDF report with recommendations.

This is an important detail. A human stays in the loop: code does not reach production without review. For teams worried that AI might quietly rewrite something critical, this format is easier to trust than fully autonomous remediation.

What This Means for Developers and Product Managers

First, OpenAI is openly entering a market where established security vendors and AI-agent startups already operate. A dedicated brand, a separate Cyber-Applied AI division, and a specialized GPT-6 Cyber model signal that this is a standalone bet rather than a side experiment.

Second, the pairing makes sense: one product finds and confirms problems, the other fixes them through the familiar code review process. If both parts perform as promised, the cycle of "find, verify, fix, test" could shrink from weeks to hours.

Now the limitations. So far, only the company's own claims are known. There are no independent tests, no figures on false-positive rates, and no data on patch quality. Pricing has not been disclosed, nor have the terms of access to Defense Factory. Red Team is in early access, so wide adoption should not be expected in the coming weeks.

The practical takeaway for teams: you can prepare in advance. Check that your project has solid test coverage, otherwise automated patches have nothing to be validated against. Decide who is responsible for reviewing these pull requests. And think through what access to your source code you are willing to give an external AI service: a tool that reads the entire repository should pass the same security review as any other contractor.

When the first independent results appear, it will become clear how close Red Team's depth comes to human testers. Until then, Daybreak remains an interesting signal: major AI labs are beginning to sell ready-made agent products for specific engineering tasks, rather than stopping at bare models behind an API.

Story

Key Questions Answered

What are the key details on Daybreak Red Team: An Automated Penetration Tester?

Red Team is currently available in early access. The task will be familiar to anyone who has commissioned a penetration test (pentest): working out how an attacker could break into a web application. The agents analyze the source code and the context of the application itself, then build a map of possible attack paths. This is easiest to explain with a building.

What are the key details on Daybreak Defense Factory: From Finding to Pull Request?

Defense Factory covers the defensive side. AI agents scan the company's codebase, look for vulnerabilities, and write the fixes themselves. The output for the developer is a ready pull request: a code change along with accompanying tests. All that remains is to review the changes and approve or reject them, and there is no need to read a PDF report with recommendations. This is an important detail. A human stays in the loop: code does not reach production without review.

What This Means for Developers and Product Managers?

First, OpenAI is openly entering a market where established security vendors and AI-agent startups already operate. A dedicated brand, a separate Cyber-Applied AI division, and a specialized GPT-6 Cyber model signal that this is a standalone bet rather than a side experiment. Second, the pairing makes sense: one product finds and confirms problems, the other fixes them through the familiar code review process.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…