OpenAI Proposes 5-Stage Plan for AI-Found Vulnerabilities
OpenAI proposed a five-stage framework for disclosing AI-found bugs after the PHP flaws, with early warnings, government coordination, and a $1B defender fund.
OpenAI has proposed a five-stage plan for how the tech world should handle serious software flaws found by AI models. The framework grew out of the company's two-month response to two critical bugs in PHP, the language that runs a large share of the world's websites. OpenAI says it expects "similar findings to increase sharply in the coming months."
How the PHP bugs led to the plan
One of the PHP flaws was found by security firm Calif using OpenAI's GPT-5.6 Sol model, as part of OpenAI's Patch the Planet initiative for open-source security. Wiz found both flaws on its own, using its Wiz Atlas platform together with Google's Gemini 4 Argon and GPT-5.6 Sol. OpenAI says that, to its knowledge, frontier AI found these vulnerabilities before attackers did - which gave defenders a head start to prepare fixes before the flaws went public.
The five stages
OpenAI's proposed sequence for flaws that need broader coordination than a normal disclosure: 1. Validate and fix upstream. Maintainers, researchers, and technical partners confirm the bug, assess its impact, and build and test a fix. Part of the testing is pointing frontier models at the patch to hunt for workarounds and related variants. 2. Give affected organizations 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 firewall rules that block malicious traffic before it reaches an application. 3. Bring in governments. For the PHP response, OpenAI and its partners engaged the White House's Super Intelligence Force, the Office of the National Cyber Director, CISA, and international partners. 4. Widen the circle. Once the fix and disclosure plan are stable, more distributors, providers, and operators join the response. 5. Publish guidance and support deployment. After patches ship, responders publish guidance and keep helping while operators deploy. The timeline is deliberately elastic. It compresses if the flaw is exploited in the wild, if details leak, if the first patch proves inadequate, or if remediation stalls in ways that raise user risk.
Why the old playbook no longer works
OpenAI is blunt about the risk: the AI tools broadly available to attackers sit close behind the frontier models doing the finding, so attackers may soon match the pace. On early-access notices, the company's own guidance is stark: "Even limited details can reveal a vulnerability." Before the PHP disclosure, Wiz partnered with Google Cloud, Cloudflare, AWS, Microsoft, Automattic, Fastly, Akamai, Imperva, Vercel, and others to prepare customer guidance, detections, and ready-made mitigations for organizations that cannot update immediately.
A $1 billion pledge for small defenders
Alongside the framework, OpenAI announced a $1 billion commitment for subsidized access to its Daybreak security tooling for under-resourced defenders, under a program it calls "Daybreak for Frontline Defenders." It also offers free Codex subscriptions for open-source maintainers and sponsors projects like the Rust Foundation.
What to watch next
OpenAI has asked the security community for feedback on the proposal, and it is clear that maintainers keep the final say on who gets warned before disclosure. The real test will come with the next AI-found flaw - and OpenAI is betting there will be many more soon.
Sources
- Help Net Security - https://www.helpnetsecurity.com/2026/10/22/openai-vulnerability-disclosure-plan/ - Unite.AI - https://www.unite.ai/php-patches-two-critical-rce-flaws-found-with-frontier-ai-models/ - OfficeChai - https://officechai.com/ai/openais-gpt-5-6-sol-helped-flag-the-critical-php-bug-flaw-thats-had-the-internet-scrambling/