PHP Fixes 2 Critical Flaws As Unpatched Servers Face Real RCE Risk

PHP maintainers fixed php2shell flaws in versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35, closing two unauthenticated remote code execution bugs that let attackers run code on exposed servers.

Key Takeaways

• PHP maintainers fixed php2shell flaws in versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 on Thursday
• The two unauthenticated remote code execution bugs are tracked as CVE-2026-109618 and CVE-2026-109619 in the U.S.

  National Vulnerability Database

• Testingcatalog named all four patched branches in an Oct. 22 post, providing defenders indicators and attackers targets to check
• Proof-of-concept publication for those CVEs would increase patch urgency because attackers would not need to build their own samples

The fixes force AI teams with older stacks into a fast patch cycle. The U.S.

National Vulnerability Database lists the two bugs as CVE-2026-109618 and CVE-2026-109619, while the flaw burst was also flagged by testingcatalog in an Oct. 22 post that named all four patched branches.

NIST’s security database tracks the same class of failure, remote code execution, which is among the most serious bug types because it can turn a normal web request into attacker-run commands on the host.

PHP Php2shell Flaws Hit The Server Layer

Unauthenticated remote code execution means the attacker may not need a valid account. A reachable service can be enough if the bug sits on the request path.

That raises the stakes for smaller AI companies.

Many ship model features in Python or JavaScript, then keep support portals, plug-in dashboards, file uploads or internal tools on older runtime services, and the model layer can stay sound while the outer application layer fails first.

The four patched branches show how broad the installed base still is. Maintainers did not fix one current line.

They issued updates for 8.2 through 8.5 because large parts of the web still run mixed versions in production.

Also Read: AMEC Eyes Korean Power Chip Wins As AI Server Racks Face Supply Risk In Trials

Old Branches Stretch The Patch Window

Maintainers issued updates for PHP 8.2 through 8.5, and defenders now need to identify where each branch is deployed before attackers can reverse engineer the fixes. Testingcatalog’s Oct. 22 post named all four patched branches, giving defenders indicators for inventory work while also providing attackers with the same targets to check.

Once patched files become public, researchers compare old and new code to isolate the exact bug path, and that diffing work can turn a fix into an exploit road map within hours.

Larger cloud operators can patch base images and rebuild services quickly, while smaller software shops often depend on old hosting panels, bundled packages or custom code that slows upgrades and creates the real exposure window.

AI Products Still Depend On Boring Plumbing

A customer support bot may call a frontier model for answers, but its login flow, upload handler or admin panel can still run on legacy runtimes. If an exposed service breaks, an attacker may reach customer data, credentials or internal tools without touching the model itself.

That is why inventory matters more than size.

A team with one good version map can move faster than a larger company that does not know which services still run 8.2 or 8.3.

Why AI Teams Need PHP Inventories

A language runtime like PHP is the software layer that executes application code on a server. When the runtime has a remote code execution flaw, hostile input can trick the server into treating attacker commands as part of the app.

That matters to AI startups because mixed stacks are normal.

A company may build inference and orchestration in Python, then keep customer accounts, dashboards or billing pages on legacy components. Integrating OpenAI or Anthropic does not remove that exposure if the web layer stays vulnerable.

The next thing to watch is exploit publication.

If proof-of-concept code appears for CVE-2026-109618 or CVE-2026-109619, patch urgency will rise again because attackers will not need to build their own working samples.

Read Next: OpenAI Helps Expose Flaws In Website Software Used By Millions