PHP2Shell Explained: The AI-Found PHP Flaws That Knocked Government Sites Offline (and What to Patch Now)

Two critical PHP flaws, CVE-2026-109618 and CVE-2026-109619 (nicknamed "PHP2Shell"), were patched and disclosed on October 21, 2026. They can let an unauthenticated attacker run code on web servers. OpenAI says its GPT-5.6 Sol model helped find one, and Wiz says it independently found both. Governments, councils and other organizations took websites offline while they patched.

What happened with the PHP vulnerabilities?

PHP's maintainers shipped emergency security releases on October 21 for two critical bugs in the language's core engine. The official PHP announcement says all PHP installations that serve web requests are at risk.

The rollout was messy. According to The Next Web, citing the Wall Street Journal, details leaked in a Hacker News post before the planned embargo ended, so PHP moved the release earlier. The SANS Internet Storm Center raised its InfoCon threat level to yellow. It cited PHP's footprint, the severity and early proof-of-concept code.

Which PHP versions are affected and which are patched?

Treat practically every PHP version as affected until you have confirmed a patched build. Wiz says the bugs reach back to PHP 5.4, and SANS says at least back to PHP 5. The fixed upstream releases are:

• PHP 8.5.12
• PHP 8.4.27
• PHP 8.3.36
• PHP 8.2.35

PHP 7.4, 8.0 and 8.1 no longer get official releases. The PHP security team did publish minimal source backports for those branches, and some Linux distributions and long-term-support vendors ship their own backported packages. Distro package numbers differ from upstream, so check each vendor's advisory. PHP 8.2 reaches end of life on December 31, 2026.

Who found the bugs, and what is the "Defender's Window"?

Several parties take credit, and the accounts don't line up perfectly.

1. OpenAI's disclosure post says GPT-5.6 Sol found one critical flaw under its "Patch the Planet" initiative. It says Wiz independently found and reported both. OpenAI calls the issue "one of the most impactful security issues in recent memory."
2. Wiz's research blog says its team used AI tooling, including Gemini 4 Argon and GPT-5.6 Sol. It says it found the first bug on August 19 and reported to PHP on August 24.
3. The security firm Calif, which OpenAI credits with using Sol, says human expertise mattered more than the model. As quoted by The Next Web, Calif wrote that no model can find "all bugs" with simple prompts.

The "Defender's Window" is OpenAI's term for the period between an AI finding a flaw and its public disclosure. Maintainers and vendors use it to prepare fixes. OpenAI says it coordinated for about two months with PHP maintainers, Cloudflare and government agencies, including CISA.

Why are critics skeptical of the AI-discovery story?

The main criticism is that the companies making the claims also sell the tools.

• Vendor framing. OpenAI and Wiz both have commercial reasons to show that AI finds bugs. Neither has published independent verification or full methods.
• Human direction. Calif's account suggests researchers chose where the model should look, so "AI found it" may overstate how independently the AI worked.
• Shared credit. A 2024 Quarkslab security audit of PHP also figures in the bugs' history. PHP maintainers wrote and shipped the actual fixes.
• Attackers get the same tools. If frontier models speed up bug discovery for defenders, they can do the same for attackers. OpenAI itself expects similar findings to rise sharply in the coming months.
• The window leaked. After the early leak, many site owners had little warning.

Who took websites offline?

According to Yle, Finland's state IT centre Valtori took the government's public web services offline as "a precautionary move." UK trade press also reported outages at some councils and at the Metropolitan Police after the NCSC urged operators to update immediately.

What should site owners do right now?

Patch first. Everything else only buys time.

1. List every system running PHP, including WordPress, Drupal, Joomla, landing-page builders and vendor appliances that bundle PHP.
2. Update to a fixed release or to your distribution's patched package. Ask hosts and agencies to confirm in writing.
3. If you can't patch today, use PHP's published interim hardening steps and web application firewall rules from providers such as Cloudflare. These are stopgaps, not fixes.
4. Watch logs for signs Wiz lists: repeated HTTP 500 errors, memory-limit errors and unexplained PHP worker restarts.
5. Consider forensic triage if a server was exposed before patching, as OpenAI recommends.
6. Plan upgrades off PHP 8.1 and older, and off 8.2 before year end.

Why does a PHP flaw matter for brand trust and AI visibility?

Most brand websites, blogs and campaign microsites run on PHP, often through WordPress. An outage takes down product pages and checkout when customers are looking. A breach can inject spam, malware or false content that search crawlers and AI systems may pick up.

AI assistants build their picture of a brand from what they can read about it. If your site is offline, defaced or tied to news about a breach, AI answers about your brand may lean on third-party coverage, outdated copies or nothing at all. We covered a similar trust dynamic in our analysis of the AI robot toy data exposure.

For marketing teams:

• Add your web stack, hosting provider and patch status to your brand-risk register, not just IT's.
• Prepare a short holding statement and status page for planned or emergency downtime.
• After an incident, check how AI assistants describe your brand. Our guide to auditing how AI describes your brand covers this.
• Keep authoritative brand facts consistent across owned and earned sources, as outlined in our GEO guide, so AI answers have reliable material even if your site is offline for a day.

FAQ

What are CVE-2026-109618 and CVE-2026-109619?

They are two critical flaws in PHP's core engine, nicknamed PHP2Shell, disclosed on October 21, 2026. PHP rates both CVSS v4.0 9.2. They can allow unauthenticated remote code execution on servers running vulnerable PHP versions.

Which PHP versions fix the vulnerabilities?

PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Older branches need vendor or source backports.

Is my WordPress site affected?

Probably, if it runs on an unpatched PHP version, because the flaws are in PHP itself and not in WordPress. Ask your host to confirm the PHP build is patched.

Did AI really find these bugs?

Partly, according to the companies involved. OpenAI says GPT-5.6 Sol found one flaw, and Wiz says it used AI tools, including Sol and Gemini 4 Argon. Calif and outside analysts say human researchers guided the work, and none of the claims has been independently verified.

Are attackers exploiting PHP2Shell?

Cloudflare reported exploitation attempts on disclosure day, and proof-of-concept code circulated after a leak. Treat unpatched, internet-facing PHP servers as at risk.

Are firewall rules enough protection?

No. PHP says no complete mitigation exists and updating is the only complete fix. Firewall rules only reduce exposure while you patch.

Last updated October 23, 2026.