OpenAI launches tools to automate security testing and patch development

Bridging the gap between bugs and fixes

OpenAI recently unveiled two new cybersecurity tools, Daybreak Red Team and Daybreak Defense Factory, during its 'Cyber at the Frontier' event. These tools are designed to expand the company's cybersecurity platform by using AI agents to investigate software vulnerabilities. The primary goal is to connect the discovery of a security flaw directly to the development of a fix.

Currently, many security teams struggle with a high volume of unverified alerts from automated scanners. These tools aim to change that by providing reproducible evidence of a weakness rather than just flagging a suspicious pattern. By validating whether an attack path is actually possible, the system helps engineers focus on real threats.

As reported by GBHackers News, the platform seeks to link offensive security testing with remediation workflows. This means the software does not just tell a company that they have a problem; it actively works to prepare the solution. This approach moves toward a model of continuous, evidence-driven assessments for web applications.

How the Red Team investigates threats

Daybreak Red Team focuses on the offensive side of security by examining web application source code and context. It uses coordinated agents to look for weaknesses and test if those weaknesses can be exploited. It can also take findings from existing bug bounty programs or vulnerability disclosures and verify them.

Nancy Wang, who leads cyber-applied AI at OpenAI, explained that the tool combines advanced penetration testing with automated scanning at scale. The objective is to ensure that engineers do not have to spend time investigating unverified alerts. Instead, the system provides the necessary context to prove a vulnerability exists.

The tool also supports retesting applications after a patch has been applied. This creates a feedback loop where the AI confirms that a fix actually works. By automating these repetitive validation steps, security teams can cover more ground in less time than manual testing allows.

Automating the remediation workflow

While the Red Team finds the problems, Daybreak Defense Factory handles the repair process. These agents investigate findings from both existing security tools and the Red Team platform to trace vulnerabilities through the code. Once a flaw is identified, the system works to develop a specific patch to fix it.

The platform is designed to work with popular tools like Wiz, Snyk, and Endor Labs, as well as GitHub repositories. For example, Endor Labs provides reachability analysis and impact information, which helps the AI agents understand the context of a finding. The system then uses policy checks to verify the resulting change before it reaches a human.

One of the key features of Defense Factory is its ability to create draft pull requests. A pull request is a way for developers to propose changes to a codebase. By packaging the remediation work into a draft, the platform allows engineers to review a finished proposal rather than starting from scratch.

The role of human oversight

OpenAI has emphasized that these tools are meant to assist security engineers rather than replace them. The platform does not automatically deploy any proposed fixes to a live environment. Instead, engineers remain responsible for reviewing and testing every change before it is merged into the final product.

This human-in-the-loop requirement ensures that the AI does not introduce new bugs or unintended behaviors into the software. The tools provide the evidence and the draft code, but the final decision rests with the human team. This maintains the traditional assurance processes that many companies require for safety.

The workflow builds on previous automation introduced through Codex Security Cloud. That system already continuously analyzes GitHub projects and prepares patches for human assessment. The new Daybreak tools represent a deeper integration of these capabilities into a unified cybersecurity platform.

Benchmarks and early access

During the product presentation, OpenAI shared some early performance data for an unreleased cybersecurity model. The model reportedly scored 77.0% on ExploitGym, which is a benchmark used to test the ability of AI to generate exploits. However, some other claims regarding the model's performance compared to other systems remain unverified.

Specifically, a chart circulating online showing a 79.1% score on PatchBench was not included on the official product pages. Because of this, those specific numbers and claims of superiority over other models are not official specifications. OpenAI has opted to keep the official documentation focused on the confirmed capabilities of the Daybreak tools.

Interested organizations can currently request early access to Daybreak Red Team. It is important to note that submitting a request does not guarantee that a company will be admitted to the program. For now, the focus remains on providing defenders with a closer connection between testing, assessment, and verification.