OpenAI has proposed a five-stage process for handling critical vulnerabilities in widely used software. The process grew out of a two-month response to flaws in PHP, the open-source language that runs websites and web applications. Calif found one of the PHP flaws using OpenAI’s GPT-5.6 Sol model, as part of OpenAI’s Patch the Planet initiative […]
OpenAI has proposed a five-stage process for handling critical vulnerabilities in widely used software. The process grew out of a two-month response to flaws in PHP, the open-source language that runs websites and web applications. Calif found one of the PHP flaws using OpenAI’s GPT-5.6 Sol model, as part of OpenAI’s Patch the Planet initiative for open-source security, and Wiz independently found and reported both.
“As significant as they are, these bugs are among several significant vulnerabilities we’ve already found and are working to fix, and we expect similar findings to increase sharply in the coming months,” OpenAI wrote.
The framework is meant for flaws that reach many products at once, where distributors and service providers need time to prepare. That time may run short if the AI tools broadly available, including to attackers, sit close behind the frontier models doing the finding.
Five stages, maintainers in charge
In the first stage, maintainers, researchers, and technical partners confirm the bug, assess its impact, and build and test a fix. Part of that testing is pointing frontier models at the patch to hunt for ways around it and for related variants.
In the second stage, affected organizations get a head start. A distributor might adapt the fix for an older release, a hosting provider might test a mitigation, and a security vendor might write web application firewall rules, which are filters that block malicious traffic before it reaches an application.
Third, responders bring in government. For PHP, OpenAI and its partners engaged the White House’s Super Intelligence Force, the Office of the National Cyber Director, CISA, and international partners. Fourth, once the fix and disclosure plans settle, maintainers widen the circle to more distributors, providers, and operators. Fifth, after a patch ships, responders publish guidance and keep helping while operators deploy.
Who hears first, and how much
OpenAI’s framework weighs three factors in deciding who gets early access: an organization’s protective role, how long it needs to act, and whether it can handle sensitive information. The same criteria would apply to comparable organizations, and maintainers make the final call. Each participant gets what its job requires, from network logs showing indicators of compromise up to working proof-of-concept exploits.
“Even limited details can reveal a vulnerability,” OpenAI wrote. Teams following the plan would revisit the timeline if the flaw is exploited in the wild, leaks, or is exposed by a revealing public patch, or if stalled remediation raises the risk to users.
What OpenAI wants help with
OpenAI is asking maintainers, researchers, and operators how long disclosure windows should run, how to weigh wide early notice against leak risk, and how to fix older package versions that no longer get security updates. One question goes further: should governments or operators use a vulnerability to patch systems they own or are authorized to manage after losing administrative access? OpenAI sketched an end state of always-on, internet-wide scanning that either alerts site owners or patches the system itself.
Download: The IT leader’s guide to AI code sprawl
