OpenAI's GPT-5.6 Sol and Google's Gemini 4 Argon models identified a critical PHP vulnerability. The discovery occurred within the Patch-the-Planet initiative, with OpenAI highlighting the flaw as a major security concern. A tool is now available on GitHub for server operators to verify if their PHP versions are impacted.
The security firm Wiz confirmed the findings using its Wiz Atlas platform in collaboration with the AI models. According to Wiz, Gemini 4 Argon assisted in creating exploits, building detection mechanisms, and validating patches. The firm noted that large language models can automatically adapt exploits to various PHP configurations, effectively lowering the barrier for scaling cyberattacks.
The remediation process spanned two months and was divided into five phases. The effort involved PHP developers, Wiz, Cloudflare, and US government agencies, including CISA and the White House. OpenAI is now proposing this collaborative framework as a standard model for addressing future security discoveries driven by artificial intelligence.
OpenAI anticipates a sharp increase in the identification of vulnerabilities similar to the PHP2Shell flaw in the near future. The company also issued a warning that the same AI capabilities used to detect security risks can be leveraged by attackers. This event contributes to the ongoing discussion regarding the dual-use nature of AI in cybersecurity.
