OpenAI proposes five-stage response plan for critical vulnerabilities

OpenAI has outlined a five-stage process for handling critical vulnerabilities in widely used software, developed during a two-month response to flaws in PHP. One flaw was found by its GPT-5.6 Sol model under the Patch the Planet initiative, and Wiz independently reported both. OpenAI expects similar AI-found bugs to rise sharply in the coming months.

• Five stages: confirm the bug, give partners a head start, bring in government, widen the circle, publish guidance after patching
• The PHP flaws were found by GPT-5.6 Sol and Wiz; OpenAI expects a sharp rise in AI-found bugs
• Early access is decided by an organization's protective role, time needed to act and handling of sensitive data
• OpenAI asks maintainers and operators about disclosure windows and leak risks from wide early notice