AI Uncovers Second Critical Vulnerability in PHP Web Infrastructure
Barely four weeks after OpenAI’s artificial intelligence identified a major security vulnerability within the PHP programming language—the backbone of approximately 70% of the internet—a second, equally dangerous flaw has been brought to light by a specialized AI system.
The PHPAfterShell Discovery
The latest discovery comes from Depthfirst, a cybersecurity firm valued at $580 million that builds proprietary models specifically for threat detection. Researchers at the company claim their AI was pivotal in isolating a critical bug, now dubbed PHPAfterShell, which mirrors the severity and exploitation potential of the previous OpenAI-discovered flaw.
The vulnerability exists within the mechanism PHP uses to manage file uploads during user sessions, such as document submissions or e-commerce transactions. By precisely interrupting an upload, an attacker can manipulate the system into executing unauthorized code, potentially leading to credential theft or unauthorized surveillance of site visitors.
The PHP Foundation is slated to release a patch tomorrow. While the Foundation’s advisory suggests that successful exploitation would typically require non-standard configurations, Depthfirst CEO Qasim Mithani maintains that the vulnerability is exploitable even under default settings.
Strategic Patching and AI-Driven Threats
Mithani advises organizations to exercise extreme caution when deploying the upcoming fix. Recent reports indicate that the National Security Agency (NSA) experienced significant system outages after attempting to patch the previous PHP vulnerability. Instead of an immediate rollout, Mithani suggests disabling the specific affected features while thoroughly testing the patch in a controlled environment.
This disclosure highlights a rapidly growing trend of AI-assisted vulnerability research. Recent developments include:
- Pentagon-contracted firm Xint using OpenAI models to identify critical flaws in NGINX.
- AI systems demonstrating increased proficiency in uncovering weaknesses across Apple’s ecosystem, including iPhones and the Vision Pro.
- Google’s report on the first instance of autonomous, AI-driven cyberattacks by Russian-speaking threat actors.
Leo Lin, a founding researcher at Depthfirst and former security expert at Apple and Snap, noted that his team utilized their AI to explore potential weaknesses in the same PHP code segments previously flagged by OpenAI. “A couple of years ago, this would have been almost impossible to find,” Mithani remarked. “A human would have spent a lot of time trying to figure this out... but now with AI, it’s become infinitely easier.”