OpenAI Warns of AI-Generated Vulnerability Report Overload

"It’s an ecosystem problem where that [maintainer] is just not going to respond to hundreds and hundreds of AI-submitted vulnerabilities," OpenAI’s cyber executive in residence, Michael Aiello, told an audience at the DataTribe Challenge in Washington, D.C.

Aiello’s warning at the DataTribe Challenge

During a question-and-answer session on Tuesday, Michael Aiello described how a surge of AI-generated vulnerability reports is overwhelming open-source maintainers, prompting some projects to stop accepting submissions altogether. Aiello contrasted that deluge with programs run by his company that take a more hands-on approach to finding and fixing flaws. He said maintainers are being asked to run through "hundreds of AI-generated reports to determine validity, the security implications and evaluate the proposed fix."

OpenAI programs and the patching tally: Patch the Planet, Daybreak, and the numbers

Aiello drew a distinction between ad hoc AI submissions into public repositories and OpenAI’s organized efforts. According to OpenAI’s website, 661 patches have been developed and tested under its initiatives, and 458 of those patches have been accepted by maintainers for inclusion in the projects they oversee. Aiello said participants in programs such as Daybreak often coordinate directly with maintainers rather than "just sending a message into GitHub and hoping for the best."

Practical advice for anyone filing a vulnerability report

Aiello offered specific, practical guidance for researchers and tools that generate vulnerability disclosures. He urged those filing reports to ensure the finding is high quality, "consumable and readable by a single person," and said reporters must "spend your own resources coordinating vulnerability disclosure." He framed that coordination as its own project: "You need to now coordinate the vulnerability disclosure across the whole ecosystem, which is itself another project."

How OpenAI handled a recent PHP vulnerability as a model

Aiello used OpenAI’s work on recent PHP vulnerabilities to illustrate the coordination he believes is necessary. He said a partner research organization discovered the flaw; OpenAI worked with the PHP Foundation to validate the issue, develop a fix and coordinate its release. The process required working with a broad group of organizations, extending "the work from technical review through patch release." Aiello acknowledged the work was imperfect: "Now, did it go perfectly? No. We’re learning as we do this over time."

What this means for open-source maintainers, security researchers, and operators

• Open-source maintainers: Faced with a high volume of AI-generated reports, some maintainers have stopped accepting submissions; those who remain will need readable, high-quality reports and are increasingly unlikely to triage large volumes without external coordination.
• Security researchers and toolmakers: Aiello’s recommendation is to elevate report quality, make disclosures consumable by a single person, and allocate resources to coordinate disclosure across affected projects—effectively taking responsibility for shepherding fixes through the ecosystem.
• Operators and response partners: The PHP example shows that responding to a vulnerability can require coordinating across foundations, maintainers and other organizations to validate, patch and issue releases—work Aiello said must be planned for and resourced.

Aiello emphasized the underlying trade-off plainly: many AI-generated reports are "likely real bugs of some degree," but maintainers must decide whether the bugs are critical and "worth the time of the maintainers." The consequence, he said, has been that some maintainers have "just stopped taking" submissions.

That blunt assessment frames a narrow but urgent policy question for those involved in the open-source ecosystem: who will do the triage and coordination work that AI tools can now generate at scale? Aiello’s answer—invest in higher-quality reporting and take on cross-project coordination—offers a concrete, if resource-intensive, path. Whether that model scales beyond organized programs like Patch the Planet and Daybreak remains the central operational challenge.

Original reporting: CyberScoop