Signal

PHP2Shell — two unauthenticated remote-code-execution flaws in PHP core (CVE-2026-109618, CVE-2026-109619, CVSS 9.2 each) — was found by GPT-5.6 Sol run by Calif under OpenAI’s Patch the Planet initiative, with Wiz independently discovering and reporting both bugs. PHP shipped fixes (8.5.12 / 8.4.27 / 8.3.36 / 8.2.35) on October 21. OpenAI’s framing: frontier AI found the bugs before attackers did, opening a “Defender’s Window” for maintainers and security providers.

Data

  • Both flaws CVSS v4.0 9.2; every PHP version from PHP 5 (2004) onward affected — PHP advisory
  • PHP powers ~70% of websites by Wiz’s count
  • Cloudflare reports in-the-wild exploitation already started; Sansec predicted mass attacks within 15 minutes of patch release
  • OpenAI’s response machinery: a five-stage coordinated-disclosure framework, engagement with the White House Super Intelligence Force, ONCD and CISA, and $1B in subsidized Daybreak access for frontline defenders
  • OpenAI says these bugs are “among several significant vulnerabilities we’ve already found” and expects findings to “increase sharply in the coming months”

Counter

Two things the coverage underplays. First, the same capability cuts both ways — OpenAI itself concedes frontier models “can also help attackers rapidly develop offensive tools,” and exploitation began within hours of disclosure; the Defender’s Window is thin, not structural by default. Second, this is partly a vendor narrative: the disclosure doubles as marketing for Daybreak enterprise access and a new Codex skill, and Wiz — the independent co-discoverer — got a fraction of the headline. Discovery credit is not a discovery monopoly.

Verdict

A mechanism-level shift, not a demo: who finds internet-scale bugs, and how the world patches them, is being rebuilt as a standing pipeline — Patch the Planet, a proposed five-stage response framework, government integration, $1B in defender subsidies. Flips down if the promised “sharp increase” in AI-found vulnerabilities fails to materialize within a few months, or if the first AI-found 0day surfaces on the attacker side first.