On October 21, 2026, Wiz Research disclosed a set of critical remote code execution flaws in PHP, the language that powers a large share of the web, alongside patched releases from the PHP project. Wiz says it found the bugs using its own Wiz Atlas system together with Google's Gemini 4 Argon and OpenAI's GPT-5.6 Sol, which makes the disclosure as much a story about AI-assisted security research as about the vulnerabilities themselves.
What was disclosed
According to Wiz, the four bugs are grouped under CVE-2026-109618 and CVE-2026-109619, and severity reaches 9.2 on the CVSS scale (the article does not give a score per CVE). Nearly every PHP version is affected by at least one issue, including releases back to PHP 5.4, and every version released in the last decade is affected by at least one. The PHP project released fixes on October 21. Wiz says the patch addresses the flaws at their core and removes the variants in the patched release, and that applying the PHP security update is the only complete remediation.
Wiz confirmed unauthenticated remote code execution against the content management systems Drupal (August 20) and WordPress (August 21) during its research, which shows the bugs could be reached through widely deployed software and not only through custom code.
Google
OpenAIThe timeline
From first discovery to patch took 63 days, and most of it was coordination
Days between disclosure milestones for the PHP vulnerabilities, from Wiz Research's published timeline.
Day counts are our calculation from the dates in Wiz Research's timeline (5 + 28 + 30 = 63 days).
| Date (2026) | Event |
|---|---|
| August 19 | First PHP vulnerability discovered |
| August 20 | Unauthenticated RCE confirmed against Drupal |
| August 21 | Unauthenticated RCE confirmed against WordPress |
| August 24 | Reported to the PHP security team |
| September 21 | PHP acknowledged the reports and began a coordinated fix |
| October 21 | Patched releases and public disclosure |
From discovery to patch is 63 days, and 58 days from report to disclosure (our arithmetic from Wiz's dates). Of that, 28 days passed between the report and PHP's acknowledgment, and another 30 between acknowledgment and patch. The discovery phase, by contrast, took five days from the first bug to the report. That ratio is the main analytical point: AI compressed the finding, but the human steps of triage, coordination across many vendors, and patch validation set the overall pace.
Where the AI models contributed
Wiz credits Gemini 4 Argon with a major contribution to discovery, exploit development, attack detection and patch validation, and credits OpenAI with help on discovery, coordination and patch verification. Raluca Ada Popa, Head of Gemini Security at Google DeepMind, said: "Using Gemini 4 Argon, the team at Wiz has shown why frontier AI is a game-changer for cyber defense."
Two cautions apply. The statements about the models' roles come from the publishers and their partners, and the disclosures do not provide a measured share of work done by AI versus humans. And finding a bug is different from knowing it is exploitable in practice, which is why the confirmed Drupal and WordPress exploitation matters.
This sits within a trend we have covered, including Gemini 4 Argon's launch and benchmarks, Anthropic's cyber verification program and an AI-built Chrome exploit chain. Wiz's earlier report on a WordPress chain, also found with GPT-5.6 Sol by another firm, shows the pattern repeating across vendors.
What defenders can do
Until patching, Wiz suggests watching for unusual multipart requests followed by memory-limit errors or repeated HTTP 500 responses, oversized or malformed PHPSESSID cookies, and unexplained PHP worker churn. Wiz notes that several vendors prepared guidance and mitigations, and that Cloudflare published WAF protections, according to AI Weekly's summary. A web application firewall is an interim control, not a substitute for the PHP update.
| Priority | Action | Why |
|---|---|---|
| 1 | Apply the PHP security release | Only complete remediation per Wiz |
| 2 | Check hosting providers and CMS vendors for bundled PHP updates | Many sites run PHP they do not manage directly |
| 3 | Add monitoring for the Wiz-listed indicators | Detect attempts before patching completes |
| 4 | Use WAF mitigations where patching is delayed | Reduces exposure in the interim |
Why the timeline matters for AI security
If AI makes discovery faster, the bottleneck moves to patch coordination and deployment. A flaw that took five days to find and 58 to disclose leaves a long window in which attackers with similar tools could also find it. That argues for faster vendor response and for treating AI-assisted discovery as a defensive priority. For teams using AI daily, including through a model-agnostic workspace like Metir, the same principle applies on a small scale: keep dependencies current, because the speed of discovery is rising on both sides.
What to watch next
- Exploitation in the wild. Whether attackers use the flaws before hosts patch.
- Patch adoption. How quickly hosting providers and CMS projects ship updated PHP.
- More AI-found disclosures. Whether further large-scope findings follow from other labs and security vendors.
- Disclosure norms. Whether acknowledgment times like the 28 days here shorten.
Sources:
- PHP2Shell: PHP RCE vulnerabilities (Wiz Research)
- AI News Today, October 21: Top Stories (AI Weekly)
- Exploitation in the Wild of wp2shell (Wiz Research)
Image credits
- Hero and in-body photo: "Heroku elePHPant (25016033688).jpg", licensed CC BY-SA 2.0, via Wikimedia Commons. Pulled 2026-10-22 and reviewed visually before use.