← All editions
Daily Brief · · 3 min read

AI found the holes. Now the web has to patch them.

The PHP security emergency shows why AI-assisted discovery needs funded maintainers, coordinated fixes and operators who can act.

ChatGPT Fan · Prepared with AI assistance

A focused technician presses one adhesive bandage across a jagged crack in a server cabinet.
AI-generated editorial cartoon · Conceptual illustration of software patching

On October 21, PHP’s maintainers released emergency fixes for two critical flaws that can let an attacker run code on a web server without logging in. AI-assisted research helped establish their severity. The immediate public benefit depends on getting the fixes installed. PHP’s advisory is the authoritative starting point.

AP reports that organizations temporarily took websites offline while updating them. That is a real operational consequence, rather than a benchmark result.

Finding a flaw is only part of the work

The PHP Foundation’s account complicates the idea that AI simply discovered something humans had never noticed. A developer found signs of one underlying problem during work following a 2024 security audit. A fix stalled amid difficult compatibility choices, an overburdened team and a decision process the Foundation calls a structural failure.

Reports from Calif, OpenAI and Wiz subsequently demonstrated the danger more concretely. The Foundation says funding let people work full time on the repair and coordination. Human maintainers made the patch decisions. Discovery, engineering capacity and distribution all mattered.

Wiz’s researchers describe AI substantially reducing the effort required to develop and adapt working exploits. Their demonstrations included WordPress, Drupal and Joomla deployments. Those are examples of affected applications, not evidence that every installation has been compromised. Avoid turning PHP’s widespread use into a claim that a fixed percentage of the entire internet has been hacked.

What a website owner can do

The maintainers list PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35 as fixed releases. Distribution vendors may supply backported fixes, so a version label alone may not settle whether a particular server is protected.

If you manage a PHP site—or pay someone to manage it—ask who owns the update, which package or hosting fix applies, and how they have confirmed the running service uses it. Updating a WordPress theme is not a substitute for fixing its underlying PHP runtime. For older installations, follow the maintainer and hosting vendor’s guidance on supported upgrades or backports.

Temporary mitigations have limits. PHP’s advisory says they do not make an installation safe; an update is the complete solution. Cloudflare describes its own protective rules, which should be read alongside the runtime update guidance.

The safety lesson

OpenAI proposes a broader coordination framework spanning maintainers, infrastructure providers and government agencies. It is a company proposal for improving response, not proof that future discoveries will be contained.

Our view: measure defensive AI by whether affected systems become safer. That means funding maintainers, reaching small operators and checking that repairs actually run. Faster discovery can help defenders while also shortening the time available to respond. It does not remove the need for controlled access, independent scrutiny or accountable decisions about increasingly capable systems.

Sources

Security advisory: Critical remote code execution vulnerabilities, update now PHP · Maintainer security advisory · Oct 21, 2026Today’s PHP Security Release, and How We’re Strengthening PHP Security The PHP Foundation · Maintainer account · Oct 21, 2026PHP2Shell: AI uncovers critical RCE vulnerabilities Wiz · Researcher account · security vendor · Oct 21, 2026Some websites go dark as operators address an AI-detected flaw AP · via ClickOrlando · Reporting · Oct 21, 2026Cloudflare deploys protections against php2shell vulnerability Cloudflare · Provider security guidance · Oct 21, 2026Patch the Planet: global response coordination as agents uncover critical vulnerabilities OpenAI · Company response proposal · Oct 21, 2026
Browse the archive →
Get in touch

Contact

Your message

Your details go to Formspree and the site owner so we can respond. They do not subscribe you to a mailing list. Privacy details

An invitation

Keep AI under human control

We love using ChatGPT. We also believe powerful AI needs independent scrutiny, meaningful safeguards, and public accountability.

We support the Statement on Superintelligence: prohibit its development until scientists broadly agree it can be built safely and kept under control, and the public supports proceeding.

You can read the statement and decide whether to add your name.

Read & sign the statement (opens in a new tab)
Hosted by the Future of Life Institute.Why we support it →