How AI Turned a Shelved 2024 PHP Bug Into a Web-Wide Emergency
PHP's maintainers spotted the root of this week's critical flaws back in 2024 but could not prove it was exploitable. Frontier AI models did that in days, and governments are now pulling websites offline.
The scariest detail in Wednesday's PHP emergency is not that artificial intelligence found a critical bug. It is that the PHP project already knew where the bug lived. In 2024, while fixing an issue from an external audit by Quarkslab, a PHP core developer spotted "the first signs" of the out-of-memory problem now tracked as CVE-2026-109618, according to the PHP Foundation's own account. A potential exploit was theorized, no concrete attack path turned up, and the fix stalled.
Two years later, frontier models supplied the attack path. That gap, between a flaw people suspect and a flaw anyone can weaponize, is what AI just collapsed.
From theory to shell in days
PHP shipped versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 on October 21 to fix two flaws, CVE-2026-109618 and CVE-2026-109619, both rated 9.2 under CVSS v4.0. The bugs sit in how PHP cleans up after fatal errors, especially out-of-memory errors, leaving stale or half-built internal state that a later request on the same worker can reuse. The result can be unauthenticated code execution or leaks of memory holding passwords and API keys.
Wiz Research, which calls the chain PHP2Shell, says every PHP release of the past decade back to 5.4 is hit by at least one of the issues. Its disclosure lays out a fast timeline. Wiz found its first bug on August 19 using its Wiz Atlas tool, Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol. It confirmed pre-authentication code execution against Drupal a day later and WordPress the day after.
"With frontier AI models and the right tools, a single researcher completed in a week what previously would have taken a world-class team months of work." Wiz Research
Wiz was not alone. The Foundation says researcher Ryan A at Calif, working under OpenAI's Patch the Planet program, reported an unconditional remote code execution with the same root cause on August 11. OpenAI independently reported the same root cause that day. OpenAI says Calif found its flaw using GPT-5.6 Sol and calls the episode "one of the most impactful security issues in recent memory."
The real bottleneck was people
The Foundation is unusually blunt about why a known weakness sat unfixed. Every candidate patch carried trade-offs, and a consensus-driven team with few engine experts could not pick one.
"We won't pretend otherwise: the delay was a structural failure, caused by both the security team's decision process and its lack of resources." The PHP Foundation
That admission reframes the AI story. Models did not invent a new class of bug. They removed the excuse that a bug was too intricate to exploit. Wiz makes the same point, arguing that as exploit work gets cheaper, "complexity may no longer provide a meaningful barrier." The Foundation adds that the team has published 15 CVEs since July 1 and waded through more than 100 invalid, duplicate or unreproducible reports. Tools can flood maintainers. They cannot choose the fix.
A disclosure that slipped its leash
The careful coordination also frayed. The Foundation says the CVE IDs leaked on October 20 with a wrong 10.0 score, pushing the release forward a day. Calif mistakenly posted, then deleted, a proof-of-concept video that spread anyway. Chainguard shipped patched images more than six hours before the embargo lifted, which the Foundation says let attackers study the fix early.
Governments reacted hard. The BBC reported that Finland temporarily shut public web services and that the NSA and the UK's National Cyber Security Centre advised taking unpatched systems offline. CISA said it had not observed exploitation at publication, but Cloudflare CTO Dane Knecht told the BBC his company was "seeing early attacks emerging."
OpenAI argues frontier AI found these flaws before attackers did, opening what it calls a "Defender's Window." That window only helps if humans can act inside it. The Foundation's own conclusion is a funding plea, not a victory lap: tools now find bugs faster than ever, "for defenders and attackers alike."
Earlier this year we covered how Claude Mythos began finding Microsoft vulnerabilities faster than they could be patched. PHP is the same lesson at internet scale. Anyone on PHP 8.1 or older gets no official release, only minimal source backports and advice to upgrade.
