If your websites run on PHP, the first thing to do about this story is patch. OpenAI says it spent two months leading a coalition of maintainers, researchers and security providers to fix critical vulnerabilities in the language that sits under a large share of the web. One flaw was found by the security firm Calif using GPT-5.6 Sol under OpenAI's Patch the Planet initiative; Wiz independently found and reported both. PHP's maintainers led the fixes and the release, and OpenAI calls it one of the most serious security issues in recent memory.
OpenAI's practical contribution is a PHP version checker released as a Codex skill, which helps operators work out whether their versions are affected and which updates apply. Behind the scenes, it says it helped test fixes, coordinate protections with hosting and infrastructure providers, and brief parts of the US government, including the Office of the National Cyber Director. It has also committed $1 billion in subsidised access to Daybreak, its programme for cyber-capable models, aimed at under-resourced organisations that protect essential services.
The larger part of the announcement is a proposed framework for handling critical flaws found by AI. It has five stages: validate and fix the bug upstream, including adversarial testing of the patch with frontier models; give affected organisations time to prepare; coordinate with government partners; prepare updates across affected products; and publish guidance once a patch exists. Maintainers decide who is told in advance, and the timeline is reassessed if exploitation, leaks or a revealing public patch appear. OpenAI says it expects findings like this to rise sharply in the coming months.
The logic is sound. When models can find serious bugs faster than small volunteer teams can fix them, disclosure needs more structure, and this mirrors what Anthropic and others have described in their own recent security work.
The governance question is harder. A company that sells the models finding the bugs is now proposing to coordinate who learns about them and when, while also deciding who gets subsidised access to those models. OpenAI asks for feedback and admits that its own blocking classifiers can get in the way of legitimate security work. The framework would carry more weight if it were owned by a neutral body such as a CERT or an open-source foundation rather than by a vendor, and the security community is right to push on that point before adopting it.