Chainguard CEO Says Suspected Exploitation Prompted Early PHP Disclosure

Chainguard CEO Dan Lorenc said he shared information early because he believed attackers were exploiting PHP flaws, putting the timing of coordinated disclosure under scrutiny.

Chainguard published images containing fixes at 3:41 a.m. ET (0741 UTC) Wednesday, Oct. 21, 6 hours and 19 minutes before PHP’s coordinated 10 a.m. ET (1400 UTC) release. The PHP Foundation said Chainguard had received the patch in advance and called the early publication an embargo breach.

The schedule had already shifted after the vulnerability identifiers began circulating around 11 a.m. ET (1500 UTC) Tuesday, Oct. 20. The PHP Foundation moved its pre-announcement to 3:30 p.m. ET (1930 UTC) that day and moved the release up one day, from Oct. 22 to Oct. 21. The Foundation said its team had not identified a functional exploit in the material it reviewed by public disclosure.

PHP fixed CVE-2026-109618 and CVE-2026-109619. Both vulnerabilities were rated Critical, with a CVSS v4.0 score of 9.2, and could allow unauthenticated remote code execution (RCE) under certain conditions. Fixed versions are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. TokenPost previously covered the PHP fixes and AI-assisted discovery.

OpenAI said Calif used GPT-5.6 Sol through its Patch the Planet initiative to identify a critical flaw. Wiz independently found and reported both vulnerabilities. The PHP Foundation said a core developer first saw signs of an out-of-memory flaw during a 2024 security audit. Calif reported an unconditional RCE path on Aug. 11, 2026, and the Foundation said OpenAI reported the same root cause that day.

Amazon said its MadPot honeypot detected an attempted exploit on Saturday, Oct. 17, four days before public disclosure. The company could not determine whether the attempt would have worked against a real system or how the actor learned about the vulnerability.

The described activity establishes an attempt against a honeypot, not a successful compromise of a live PHP server. The AI work described by OpenAI and the PHP Foundation concerned vulnerability research; it does not establish that AI was used in the honeypot attempt. PHP maintainers reviewed the reports and fixes.