Frontier AI models helped expose 'PHP2Shell' flaws in core PHP; OpenAI proposes a five-stage disclosure plan

On 21 October 2026 the PHP project released patched versions fixing memory-safety flaws in the language’s core engine, and Wiz Research publicly disclosed them under the name PHP2Shell.

The bugs

• The issues are grouped under two CVEs with severity reaching CVSS 9.2. Wiz says they can allow unauthenticated remote code execution and disclosure of server memory that may hold passwords, tokens and API keys.
• According to Wiz, every PHP version back to 5.4 is affected by at least one of the flaws. Wiz demonstrated working attacks on WordPress, Drupal and Joomla, but stresses the bugs are in PHP itself, not in those applications. Wiz says PHP powers about 70% of websites.
• The only complete fix is to update PHP. Cloud and CDN providers prepared temporary mitigations for those who cannot patch at once.

The role of AI

Wiz says it used its own tool Wiz Atlas plus Gemini 4 Argon and GPT-5.6 Sol, and that the models made it practical to chain the flaws into exploits in hours, work it calls “at the edge of what human researchers could practically achieve”. Separately, a researcher at the firm Calif found one of the flaws with GPT-5.6 Sol under OpenAI’s Patch the Planet open-source security programme.

OpenAI’s disclosure plan

OpenAI used the case to propose a five-stage process for AI-found bugs that hit many products at once: confirm and fix (including pointing frontier models at the patch to hunt for bypasses), give distributors and hosts a head start, involve government responders, widen the circle, then publish guidance. For PHP, Help Net Security reports, this included the Office of the National Cyber Director and CISA. OpenAI says it expects such findings to “increase sharply in the coming months” and is asking the security community for feedback.