Web Security: AI finds critical PHP flaws affecting 70% of sites

Seventy percent of the internet runs on PHP, a fact that became central to a major cybersecurity event this week after OpenAI announced that its AI model had helped uncover critical security flaws in the language. These vulnerabilities, which had eluded human developers for two years, were turned into working exploits by AI-assisted researchers in mere weeks. The discovery triggered a coordinated response involving PHP maintainers, security providers, and government agencies to patch the issues.

The flaws, tracked as CVE-2026-109618 and CVE-2026-109619, stem from how PHP manages computer memory and handles errors. One issue occurs when the system runs out of allocated memory, potentially leaving incomplete data that the engine later treats as valid. The second involves a fatal error causing the system to point to memory that has already been freed, allowing subsequent requests to access or corrupt it. Both issues received a critical severity score of 9.2 out of 10. Because these bugs exist in the core engine, websites can be vulnerable even if their specific application code is secure.

The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 on October 21, bringing the release forward from its original October 22 schedule due to premature disclosures on October 20. Users of PHP 8.1 and older versions will not receive official fixes. OpenAI notes that any site using PHP can safely upgrade to remain secure. The vulnerabilities allow remote attackers to execute malicious code without login credentials, potentially taking control of servers and exposing sensitive data such as passwords, authentication tokens, and API keys stored in server memory.

The work was conducted under OpenAI’s “Patch the Planet” initiative, with researchers from the cybersecurity firm Calif beginning audits on July 30, 2026, at the PHP Foundation's request. The Foundation had known about the underlying flaw since 2024 but could not identify an exploitation method. Calif researcher Ryan A. first identified the exploitable vulnerability on August 11, followed by OpenAI researchers the same day. Another firm, Wiz, independently identified an additional vulnerability. While human teams had examined the code, they could not fully resolve the issues, allowing them to persist for two years.

OpenAI describes this incident as an example of the “Defender's Window,” a concept introduced after its models hacked Hugging Face. This refers to a narrow timeframe where organizations can use advanced AI to find and patch defenses before those same capabilities become widely available to malicious actors. The company argues that it is a collective responsibility to act proactively, noting that Anthropic’s Claude Mythos recently identified thousands of zero-day vulnerabilities in major operating systems and browsers.

The lengthy two-month timeline to fix the issues was attributed to the complexity of the task, which required major changes to the PHP engine. While some may argue the flaws were effectively harmless because they would not have been exploitable without AI, OpenAI challenges this view, emphasizing that powerful technology will inevitably be used by bad actors. The company considers this one of the biggest cybersecurity discoveries of 2026 given the ubiquity of PHP.

Going forward, web developers and system administrators are urged to update their PHP versions to the latest patched releases. The incident highlights the growing role of AI in both identifying and exploiting software weaknesses, shifting the dynamic of digital security defense.