SI Agent News
DevelopingClaudeOpenAIAgentsSecurity

OpenAI launches AI agents to find and patch web app vulnerabilities

The new Daybreak suite uses GPT-6 Cyber to automate penetration testing and generate code fixes while keeping humans in control of the final deployment.

Papercut shadow box: picture for the story 'OpenAI launches AI agents to find and patch web app vulnerabilities'
AI-made papercut shadow box for this story. No real people or products are shown.
948vulnerabilities found in Linux-Kernel by GPT-6 Cyber

The short version

Why it matters

This marks a shift toward automated 'self-healing' software where AI does the heavy lifting of finding and fixing bugs. For companies, it means faster security updates, but it also highlights the growing role of AI in the cyber warfare arms race.

Introducing the Daybreak suite

OpenAI recently unveiled two new security products, Daybreak Red Team and Daybreak Defense Factory, during its 'Cyber at the Frontier' event. These tools are part of a broader platform designed for verified defenders who need to secure their digital infrastructure. The primary goal of these products is to identify security flaws in web applications and source code before malicious actors can exploit them.

The new suite is built to handle the entire lifecycle of a security bug, from the initial discovery to the final repair. By using specialized AI agents, OpenAI aims to make the process of finding and fixing software vulnerabilities more efficient. These tools are intended to assist security teams rather than replace them entirely, ensuring that humans remain the final decision-makers in the security chain.

As it-daily.net reported, these products are currently being rolled out in different stages of availability. While Daybreak Red Team is entering an early access phase for selected customers, Defense Factory is currently available as a research preview. This phased approach allows OpenAI to gather feedback from professional security teams while refining the underlying technology.

Automated penetration testing

The Daybreak Red Team service functions as an automated penetration tester. In cybersecurity, a penetration test is a simulated attack used to find weaknesses in a system. This specific tool uses multiple coordinated AI agents to examine web applications by analyzing their source code and the broader context of how the application is used.

These agents do not just scan for known issues; they are designed to play out various attack paths. They can also verify reports coming from other sources, such as bug bounty programs or standard security scanners. A key feature of this tool is its ability to test whether a specific patch actually prevents a known exploit from working, providing a layer of verification for developers.

Johannes Landgraf, who manages cybersecurity at OpenAI, noted that the Red Team tool is intended to expand access to their frontier cyber models. The company wants to provide these advanced capabilities to a wider range of users over time. For now, interested parties must join a waitlist, though joining the list does not guarantee immediate access to the service.

From discovery to pull requests

While the Red Team finds the problems, the Daybreak Defense Factory is responsible for fixing them. This tool takes findings from existing security tools or the Red Team and traces the specific vulnerabilities through the application's code. It then attempts to recreate the flaw in an isolated environment to confirm its behavior before proposing a solution.

Once a vulnerability is confirmed, the AI agents prepare a patch and a corresponding test for that specific bug. The final output is a draft for a pull request, which is a standard way for developers to submit code changes to a repository. This means the AI does not have the authority to change the live software on its own; a human developer must review the code and manually approve the change.

Defense Factory is currently accessible via the Daybreak plugin in ChatGPT and Codex. It works with connected GitHub repositories, though it is currently restricted to personal accounts. This setup allows developers to see how AI can assist in the tedious work of writing and testing security fixes without risking their entire production environment.

The power of GPT-6 Cyber

The underlying engine for these new tools is the GPT-6 Cyber model. OpenAI introduced this model on November 5, highlighting its specific capabilities in the cybersecurity domain. According to the company, the model has already demonstrated its effectiveness by identifying 81 severe vulnerabilities in Firefox and 948 vulnerabilities in the Linux-Kernel.

The performance of these models is a major point of interest for enterprise security. OpenAI President Greg Brockman stated during the event that the security of organizations increasingly depends on which frontier models they choose for defense and how they implement them. This suggests that the choice of AI model is becoming a strategic decision for companies protecting sensitive data.

OpenAI is not alone in this space, as other major tech companies are developing similar AI-driven defense mechanisms. Google has expanded its CodeMender agent to perform attack simulations in sandboxes, and Anthropic offers a service called Claude Security for continuous code auditing. The competition is creating a rapidly evolving landscape where AI agents are becoming standard tools for maintaining software integrity.

What to watch

Try it

If you use GitHub, check if you can access the Daybreak plugin via your personal account to see how it interacts with your repositories.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by it-daily.net · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news