PHP2Shell: PHP patches two critical unauthenticated RCE flaws (CVSS 9.2) found with OpenAI's GPT-5.6 Sol and by Wiz; OpenAI led coordinated disclosure and pledges $1B in subsidized Daybreak access

Why it matters

【Key points】

PHP shipped emergency fixes at 14:00 UTC on Oct 21 for two unauthenticated RCE flaws in the core engine ("PHP2Shell"), each CVSS v4.0 9.2. OpenAI says Calif found a critical flaw with GPT-5.6 Sol under Patch the Planet; Wiz independently found both.

1. CVE-2026-109618 (OOM interrupts partially initialized request data) and CVE-2026-109619 (stale request state after free). All web-serving PHP is affected, mod_php worst; Cloudflare says PHP 5.4+.

2. Fixed in 8.5.12 / 8.4.27 / 8.3.36 / 8.2.35; source backports for 7.4–8.1.

3. Cloudflare WAF blocked by default from 03:02 UTC and saw attack traffic after a pre-disclosure leak. OpenAI engaged the Super Intelligence Force, ONCD and CISA and proposed a five-stage response framework.

4. OpenAI adds a version checker, an upcoming Codex skill and $1B in subsidized Daybreak access, and expects such findings to "increase sharply".

【Verifiable】

https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/

https://www.php.net/archive/2026.php

https://blog.cloudflare.com/php-vulnerabilities-waf-protection/

https://www.wiz.io/blog/php2shell-php-rce

【Our read】

1. Background: first public case of a frontier model finding internet-scale flaws in core web infrastructure (~70% of sites).

2. Comparison: Anthropic's same-day attack-surface study (493 issues in 25 projects) shrinks reachable code; Google paused OSS bug-bounty intake over invalid AI reports. OpenAI ran an end-to-end coordinated disclosure.

3. Impact: patch now, especially mod_php. A leak before disclosure shows embargo windows are shrinking.

4. Caveats: "found before attackers" is OpenAI's claim; check Calif/Wiz write-ups for GPT-5.6 Sol's exact role.