PHP2Shell brought a familiar security challenge into sharper focus: vulnerabilities that were once difficult to exploit can become more practical as AI tools speed up analysis. Millions of attempts to use the newly public PHP flaws were recorded within hours, according to Cloudflare analysis. The PHP Foundation said it had not seen successful exploitation, and many circulating exploit attempts appeared ineffective.
Old issues became newly actionable
The PHP2Shell vulnerabilities were identified by researchers at Calif, who developed exploits in early August. Their work used frontier AI tools through OpenAI’s Patch the Planet program alongside specialist human knowledge of PHP and its weaknesses.
Some of the underlying issues had been identified in 2024, but were considered too complex to exploit at that time. After Calif’s discovery, researchers at OpenAI and the Google-owned security firm Wiz independently found PHP2Shell flaws using AI tools.
Wiz security researcher Sagi Tzadik said the vulnerability had been known to the PHP team for years, but had not been addressed because an attacker could not leverage it, at least not a human attacker. The change was not that the underlying weakness suddenly appeared; the practical path to exploiting it had changed.
AI speeds up a familiar security race
AI models can make it faster to examine software and identify ways a flaw might be exploited. That speed matters after vulnerabilities become public, when defenders are trying to understand the risk and update systems while attackers are also looking for working methods.
Cloudflare’s report of millions of attempts shows how quickly attention turned to PHP2Shell. But the attempts did not establish that attackers had succeeded. The PHP Foundation reported no successful exploitation, and potential exploits circulating publicly appeared ineffective.
The distinction matters: a surge in probing can create urgency for web administrators without proving that a breach has happened. The immediate challenge is to respond to disclosed flaws while separating attempted exploitation from confirmed compromise.
People remain part of the discovery process
The researchers’ accounts do not describe AI working alone. Calif’s discovery paired frontier tools with human expertise in PHP. Ryan Austin, the Calif researcher who first found the flaws, wrote that no model can uncover all bugs through simple prompts.
Security researchers say advanced AI is amplifying established patterns in vulnerability research and coordinated disclosure. It can help skilled teams move faster, but the PHP2Shell case still required a highly capable team to identify and develop practical exploits.
That context tempers predictions of an immediate flood of newly exploitable historic bugs. Jake Williams, vice president of research and development at Hunter Strategy and a former NSA hacker, expects more long-lived bugs to be found across software versions. He also said PHP’s internet exposure makes its attack surface unusual compared with tools that are not designed to be exposed online.
Pressure on defenders, without a reason to panic
The workload for security teams is already substantial. Microsoft’s security updates included a record 1,453 security vulnerabilities this month alone. Bank of America recently hired 1,000 consultants to fix flaws discovered by Anthropic’s Mythos.
Those examples point to a practical concern: as AI helps researchers analyze flaws faster, under-resourced IT teams, developers, and administrators face pressure to keep pace. Organizations need to account for how quickly a hard-to-exploit weakness can become a more immediate concern.
Still, the PHP2Shell episode is not proof that AI will expose every old bug or make every vulnerability easy to exploit. Security engineer and researcher Niels Provos said he was not particularly worried about an overwhelming wave, noting that the discovery required a highly skilled team. For defenders, the lesson is to take disclosure and patching seriously while judging risk by evidence, not by the volume of attempted attacks alone.