Back to the stories

OpenAI says frontier AI helped identify critical PHP vulnerabilities before public disclosure

Score 10.0,

AI cybersecurity

Frontier AI models helped find critical security flaws in PHP before those flaws became public, and the problem was fixed through coordinated action by maintainers, private security teams and agencies.

OpenAI says a team using GPT-5.6 Sol, as part of its Patch the Planet work, identified a critical PHP flaw. OpenAI also says Wiz independently discovered and reported both vulnerabilities and coordinated with PHP maintainers and other security providers. Reuters reported that Wiz used its Wiz Atlas autonomous research tool alongside Google's Gemini 4 Argon and GPT-5.6 Sol.

Maintainers, OpenAI, Wiz and EU cybersecurity agencies spent a two-month coordination period preparing patches and protections, and the vulnerabilities were addressed before public disclosure. Agencies warned some organisations to consider taking affected sites offline while fixes were staged.

Why this matters: PHP runs a large share of the web. A single critical bug there can expose thousands of sites. The new element is that frontier models are now helping security teams find those bugs faster, which can be a net win for defenders if discoveries are handled responsibly.

How it works: think of these models as very fast code-review assistants. They can scan patterns across huge codebases and flag unusual or risky constructions, but a skilled human still has to validate the finding and coordinate a safe fix.

What changes now: security teams should expect AI-assisted discovery to become part of their toolset, and software maintainers and regulators will need clear playbooks for coordinated disclosure. Running this kind of research and managing fixes still requires multi-organisation expertise and careful oversight, not just an off-the-shelf model.

The open question is whether coordinated disclosure processes can scale as AI finds more and faster issues across widely used infrastructure.