PHP patches two critical flaws
On 21 October 2026 the PHP project released versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to fix two critical vulnerabilities, CVE-2026-109618 and CVE-2026-109619, each rated 9.2 on CVSS 4.0, that let an unauthenticated attacker run code on a server. The same day OpenAI published its own account of a two-month coalition response under its Patch the Planet initiative, in which, it says, the security firm Calif found a flaw using GPT-5.6 Sol, and committed $1 billion in subsidized Daybreak access to defenders.
Why it matters
The case can be checked against the maintainers' own account, not only the lab's. The PHP Foundation says the out-of-memory weakness was first glimpsed in a 2024 audit but its fix stalled for months over trade-offs, until reports from Calif (11 August), Wiz (24 August) and OpenAI made the full impact clear; it also names AI tools it now uses to validate reports. OpenAI calls the flaws one of the most impactful security issues in recent memory and proposes a five-stage framework for responding to findings of this kind; that framing is OpenAI's own. Singapore's cyber agency reported widespread exploitation attempts by 23 October.
The release. The php.net advisory of 21 October says both flaws can allow unauthenticated remote code execution, that all PHP installations serving web requests are at risk (the highest risk is with Apache mod_php, where a demonstrated exploit needs no application code), and that there are no complete mitigations: updating is the only safe choice. CVE-2026-109618 is an out-of-memory handling flaw that leaves request data partly initialized (GHSA-w8hw-w9fx-j79w); CVE-2026-109619 is request state that survives after its memory is freed (GHSA-6wmf-4rv5-p3rp). The Cyber Security Agency of Singapore (alert AL-2026-151, 21 October, updated 23 October) calls the pair PHP2Shell and says all versions of PHP are affected. How it was found, per the PHP Foundation (post of 21 October). A 2024 external audit of PHP, the first in years (Quarkslab, funded by the Sovereign Tech Agency), gave the first signs of the memory problem; approaches to a fix stalled over compatibility and denial-of-service trade-offs, and the Foundation calls the delay "a structural failure". Calif began an audit under Patch the Planet on 30 July; on 11 August it reported an unconditional remote code execution with the same root cause; Wiz reported further vectors on 24 August; OpenAI independently reported the same root cause and sent further vectors on 18 September. The Foundation says it uses OpenAI's Daybreak Red to validate reports and exploits and earlier scanned PHP with Anthropic's Claude Mythos. The disclosure. The CVE identifiers leaked on 20 October at about 15:00 UTC with a wrong CVSS score of 10.0; the release was moved from 22 October to 21 October, 14:00 UTC; Calif mistakenly published a proof-of-concept video; Chainguard published fixed images before the release, which the Foundation calls a breach of the embargo. OpenAI's post (21 October). OpenAI says it led a coalition for two months, that Calif identified a critical flaw using GPT-5.6 Sol with Wiz independently reporting both, that it shares a PHP version checker as a Codex skill, and that it committed $1 billion in subsidized Daybreak access through "Daybreak for Frontline Defenders". It proposes five stages: validate and fix upstream; give affected organizations time to act; coordinate with government partners (the post names the White House's "Super Intelligence Force", ONCD and CISA); prepare updates across products; publish guidance. "To our knowledge, frontier AI found these vulnerabilities before attackers did" is OpenAI's claim. What the record does not claim. OpenAI's post gives no CVE number, version or release day, and the Foundation does not say which model Calif used. Who found the flaws first, and the share of AI in it, rest on the parties' own accounts. The $1 billion is a commitment, not a measured outlay.