Critical PHP Flaws Exposed: AI-Driven Discovery Forces Rapid Patch for Two-Thirds of the Web

OpenAI announced on Tuesday that its GPT-5.6 Sol model successfully identified critical security vulnerabilities in PHP, a programming language that powers an estimated 70% of the internet. The discovery, made under the company’s “Patch the Planet” cybersecurity initiative, resulted in the emergency release of patched software versions to protect a vast portion of global web infrastructure from potential remote attacks.

The two flaws, tracked as CVE-2026-109618 and CVE-2026-109619, stem from errors in how PHP manages computer memory and processes fatal errors. The first issue occurs when the system runs out of allocated memory during a request, leaving incomplete data that the engine may later misinterpret as valid. The second involves a fatal error that leaves the system pointing to freed memory, allowing subsequent requests to access or corrupt that space. Both vulnerabilities carry a critical severity score of 9.2 out of 10.

Researchers from Calif, a California-based cybersecurity firm and founding partner of the Patch the Planet program, began auditing PHP with OpenAI tools on July 30, 2026, at the request of the PHP Foundation. While the Foundation had been aware of the underlying flaw since 2024, it had been unable to determine an exploitation method or a fix. On August 11, Calif researcher Ryan A. identified the exploitable vulnerability, a finding that was corroborated the same day by OpenAI researchers. Independently, the cybersecurity firm Wiz identified an additional related vulnerability.

The coordinated release of fixes was originally scheduled for October 22 but was accelerated to October 21 following premature disclosures on October 20 that triggered public concern. The PHP Foundation has released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the issues. Users running PHP 8.1 or older versions will not receive official fixes, prompting OpenAI to advise that all sites using PHP can safely upgrade to the newer versions to remain secure.

OpenAI describes this incident as a prime example of its “Defender's Window” concept, a theory introduced after its models hacked Hugging Face. The concept posits that there is a narrow timeframe during which organizations can use advanced frontier AI models to find and patch defenses before those same capabilities become widely available to malicious actors. The company argues that while the flaws might have remained theoretical without AI, the technology’s inevitability creates a collective responsibility to act proactively against potential threats.

The discovery highlights a significant shift in cybersecurity dynamics, where AI-assisted researchers turned a two-year-old bug that stumped human developers into working exploits within two weeks. OpenAI notes that human teams had examined the code for a long period without fully resolving the issues, allowing them to persist until the AI-driven audit. This event is being viewed as one of the most significant cybersecurity discoveries of 2026 due to the widespread use of PHP.

The implications extend beyond simple code errors, as the AI-assisted researchers demonstrated that the flaws could expose sensitive data stored in server memory, including passwords, authentication tokens, and API keys. Because the vulnerabilities exist in PHP’s core engine, websites can be affected even if their specific application code is secure. Attackers can potentially exploit these weaknesses remotely without needing login credentials to execute malicious code and take control of servers.

OpenAI is challenging the view that such flaws are harmless if they are only exploitable with advanced AI. The company points to recent findings by Anthropic’s Claude Mythos, which identified thousands of zero-day vulnerabilities in major operating systems and web browsers, to support its argument for proactive defense. The incident underscores the growing role of artificial intelligence in both creating and mitigating risks in the digital landscape.