AI frontier models from OpenAI and Google have identified a major security vulnerability in PHP that impacts millions of websites worldwide. OpenAI’s GPT-5.6 Sol uncovered the flaw through its Patch-the-Planet security initiative, while cybersecurity firm Wiz independently utilized Google’s Gemini 4 Argon alongside OpenAI’s model on its Wiz Atlas platform to discover the bug, write functional exploits, and validate patches. The findings were addressed via a coordinated, two-month remediation process involving PHP maintainers, Cloudflare, CISA, and the White House.
Wiz highlighted that frontier LLMs radically transform vulnerability exploitation by automatically adjusting attack payloads across varying PHP versions, operating systems, and memory configurations—dramatically lowering the technical barrier required to execute broad, multi-target attacks. While OpenAI framed the advance warning as a landmark success for AI-driven defense, security researchers caution that these dual-use capabilities will similarly accelerate offensive exploit generation by bad actors in the coming months.
Why it matters
Proves frontier LLMs can independently identify zero-day vulnerabilities in core legacy web infrastructure at massive enterprise scale.
Lowers the technical barrier for threat actors by automatically adapting exploit payloads across custom server memory configurations.
Establishes a new blueprint for coordinated disclosure between frontier AI labs, security vendors, and federal agencies.
Source: the-decoder.com



