Testing a PHP2Shell exploit in the era of AI

Reporting by Intel 471 BlogRead the original at intel471.com

Executive Summary

Critical memory-safety flaws, CVE-2026-109618 and CVE-2026-109619, were discovered in the core PHP interpreter by researchers utilizing frontier AI models. These vulnerabilities are rated with a CVSS v4.0 score of 9.2 and affect nearly 70% of websites built on PHP, alongside embedded systems. Although initial reports saw a flood of non-functional Proofs of Concept (PoCs) attributed to AI hallucinations, highly skilled researchers later developed working exploits demonstrating Remote Code Execution (RCE). Testing by Intel 471 showed that while the exploit can achieve RCE against specific production configurations—especially in Apache modphp environments without additional application vulnerabilities—reliability remains a significant hurdle, succeeding only once out of two dozen lab attempts. This contrasts with PHP-FPM implementations, which offer certain defensive postures but still present exploitable paths under controlled circumstances.

Facts Only

Critical memory-safety flaws CVE-2026-109618 and CVE-2026-109619 exist in the core PHP interpreter.
The vulnerabilities affect approximately 70% of websites, including WordPress, Drupal, and Joomla.
Exploitation attempts began several days before public disclosure on October 21.
CISA warned that these vulnerabilities are being actively exploited in the wild.
A prolific ransomware affiliate claimed to possess a working exploit.
Testing reproduced unauthenticated RCE against an "php:8.4.26-apache" Docker image using modphp configurations.
CVE-2026-109618 relates to memory corruption arising from PHP's handling of OOM conditions via "longjmp()".
CVE-2026-109619 is a use-after-free vulnerability related to how PHP retains references to request-specific memory.
A suspected China-linked actor chained PHP2Shell with a Linux kernel privilege escalation flaw on October 26 in the US, Japan, Taiwan, and South Korea.
Official security updates are recommended for systems running older PHP versions.

Full Take

The narrative surrounding PHP2Shell illustrates a critical friction point between theoretical vulnerability severity and practical operational risk. The initial panic stemmed from two powerful forces converging: high-impact zero-day flaws in ubiquitous software (PHP) paired with the emergent threat of AI accelerating attack development, leading to reports of "AI-hallucinated PoCs." This juxtaposition prompts us to question what constitutes actionable intelligence—is a claimed exploit merely sophisticated noise until it survives hostile environmental testing? The finding that reliable exploitation is severely environment-dependent—struggling due to ASLR bypass and heap layout variance even across identical builds—challenges the notion of an immediate, widespread catastrophic threat like Log4Shell.
This scenario reveals a pattern wherein advanced capability (AI research enabling sophisticated bug discovery) outpaces operational hardening capability in diverse production environments. While the high CVSS scores demand attention, the actual barrier to mass exploitation is currently situated in implementation variance and low reliability rather than purely theoretical cryptographic weakness. Furthermore, the divergence between modphp susceptibility (easier application-independent attack paths) and PHP-FPM defenses underscores that generic patching guidance masks essential architectural context; risk is deployed unevenly across infrastructure stacks. The push for immediate remediation is justified, not solely based on current exploitability statistics, but on preempting the inevitable evolution of AI tooling capable of resolving these environmental variability challenges at scale.
Patterns detected: none
Autonomy check: No autonomy signals.

From the original · Intel 471 Blog

hallucinated PoCs Intel 471 tested a public exploit for the AI-discovered PHP flaws. We found exploitation is hard, but you should still patch now.

Read the full story at intel471.com

Sentinel unavailable

The automated check of the source article's wording did not complete, so no result is shown.

This looks only at the wording of the original source article, not at this page's AI-written sections. A small local AI model made this estimate. It has not been checked against known human and machine texts, so treat it as provisional. It cannot show who wrote an article.