AI Models Expose Critical PHP Flaws Affecting Seven-Tenths of the Web

A widespread security threat to the modern web has been identified and patched after AI researchers found a way to exploit long-hidden flaws in PHP.

OpenAI announced on October 21 that its GPT-5.6 Sol model helped uncover critical vulnerabilities in the programming language that powers an estimated 70% of the internet. The discovery was made under the company's 'Patch the Planet' cybersecurity initiative. The flaws allow attackers to execute malicious code remotely without needing login credentials. This can lead to full control of affected servers and exposure of sensitive data like passwords and API keys.

The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the issues. Older versions, including 8.1 and below, will not receive official fixes. The release was originally planned for October 22 but was moved up to October 21. This change occurred after premature disclosures on October 20 caused public concern. OpenAI coordinated a coalition of researchers, maintainers, and government agencies to resolve the issues over two months.

The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, involve errors in how PHP manages memory and handles faults. One flaw occurs when the system runs out of memory, potentially leaving invalid data that the engine mistakes for valid information. The other involves a fatal error leaving the system pointing to memory that has already been freed. Both issues received a critical severity score of 9.2 out of 10. Because the bugs exist in the core engine, sites are vulnerable even if their specific application code is secure.

Researchers from Calif, a founding partner of the Patch the Planet program, began auditing PHP on July 30. The PHP Foundation had known about the underlying flaw since 2024 but could not find a way to exploit or fix it. Calif researcher Ryan A. identified the exploitable vulnerability on August 11, with OpenAI researchers doing the same that day. Another firm, Wiz, independently found a second vulnerability. While human developers struggled to resolve the code issues for two years, the AI-assisted team demonstrated practical exploits within two weeks.

OpenAI framed this incident as a prime example of its 'Defender's Window' concept. The company argues that there is a narrow timeframe where organizations can use advanced AI to find and patch flaws before attackers do. This follows earlier instances where OpenAI models hacked Hugging Face. The firm maintains that powerful technology will inevitably be used by bad actors, making proactive defense a collective responsibility. This view counters the argument that AI-assisted bugs are harmless if they require AI to exploit. Anthropic recently identified thousands of similar zero-day vulnerabilities in major operating systems and browsers.