OpenAI’s AI Model Exposes Critical PHP Flaws Affecting Most of the Web

OpenAI has revealed that its GPT-5.6 Sol model, deployed under the company’s Patch the Planet cybersecurity initiative, identified critical security weaknesses in PHP, the open-source language estimated to underpin 70% of the internet. The announcement, made on Oct. 21, highlights a significant shift in how vulnerabilities are discovered, as AI-assisted researchers converted a bug that had stumped human developers for two years into functional exploits within just two weeks. The company coordinated with PHP maintainers, security providers, and government agencies to address the issues, attributing the two-month remediation period to the complexity of altering the core PHP engine.

The scale of the fix is immediate and specific. The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 on Oct. 21, moving the coordinated release forward from its original Oct. 22 schedule after premature disclosures sparked public concern on Oct. 20. Users of PHP 8.1 and older versions are left without official fixes, though OpenAI states that all sites can now safely upgrade to secure their infrastructure. Both vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, carry a critical severity score of 9.2 out of 10.

The discovery process involved a coalition of technical experts. Researchers from Calif, a California-based cybersecurity firm and founding partner of the Patch the Planet program, began auditing PHP on July 30, 2026, at the request of the PHP Foundation. Ryan A., a Calif researcher, first identified the exploitable vulnerability on Aug. 11, with OpenAI researchers confirming the finding that same day. Wiz, another cybersecurity firm, independently identified a second vulnerability during the same period. The PHP Foundation had been aware of the underlying flaw since 2024 but lacked the method to exploit or fix it until the AI intervention.

The technical stakes are high. The flaws involve errors in how PHP manages computer memory and handles exceptions. One occurs when the system runs out of allocated memory, leaving incomplete data that the engine later treats as valid. The second involves a fatal error leaving PHP pointing to already freed memory, allowing subsequent requests to access or corrupt it. Attackers can exploit these weaknesses remotely without login credentials to execute malicious code, potentially taking control of websites. The AI analysis also demonstrated that these bugs could expose sensitive data, including passwords, authentication tokens, and API keys, stored in server memory.

OpenAI frames this incident as a practical example of its “Defender’s Window” concept, a period during which organizations can use advanced AI to strengthen defenses before malicious actors do. The company argues that since powerful technology will inevitably be used by bad actors, proactive patching is a collective responsibility. This stance challenges the view that these flaws were harmless because they required AI to become exploitable. The company cites a recent discovery by Anthropic’s Claude Mythos, which identified thousands of zero-day vulnerabilities in major operating systems and browsers, as further evidence of this shifting security landscape.

For New England businesses and developers, the urgency is clear. If your infrastructure relies on PHP, the window for safe operation has narrowed. While the patches are now available, the incident underscores a growing dependency on AI-driven security audits to catch what traditional human review missed. The two-year gap between the flaw’s existence and its exploitation highlights a persistent risk in legacy systems, one that may become increasingly common as AI tools become more widespread in both defense and offense.