AI models uncover deep flaws in PHP affecting majority of web

OpenAI revealed on Monday that its AI model, GPT-5.6 Sol, assisted researchers in identifying critical security weaknesses within PHP, the open-source programming language that powers an estimated 70% of the internet. The discovery, made under the company’s “Patch the Planet” cybersecurity initiative, led to a coordinated response involving PHP maintainers, security providers, and government agencies. The company stated that the complexity of the issues, which required significant changes to the PHP engine, necessitated a two-month period to develop and distribute fixes.

The PHP Foundation has issued patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the vulnerabilities. Users of PHP 8.1 and older versions will not receive official updates. Although the coordinated release was originally planned for today, October 22, it was moved up to yesterday, October 21, after premature disclosures on October 20 sparked public concern. OpenAI advised that all sites utilizing PHP should upgrade their versions immediately to ensure security.

The initiative began on July 30, when researchers from Calif, a California-based cybersecurity firm and founding partner of the program, started auditing PHP at the request of the PHP Foundation. The Foundation had been aware of the underlying flaw since 2024 but had been unable to determine a method for exploitation or a viable fix. On August 11, Calif researcher Ryan A. identified an exploitable vulnerability, a finding that was mirrored by OpenAI researchers the same day. Shortly after, another firm, Wiz, independently discovered an additional flaw.

The two vulnerabilities, identified as CVE-2026-109618 and CVE-2026-109619, stem from errors in how PHP manages memory and processes errors. The first issue arises when PHP exhausts its allocated memory during a request, potentially leaving incomplete internal data that the engine later incorrectly interprets as valid. The second flaw occurs when a fatal error causes PHP to reference memory that has already been freed, allowing subsequent requests to access or corrupt that data. In specific scenarios, attackers can exploit these weaknesses remotely without needing login credentials to execute malicious code and potentially seize control of affected servers.

Using AI tools, researchers demonstrated that these flaws could also expose sensitive data stored in server memory, such as passwords, authentication tokens, and API keys. The vulnerabilities reside in PHP’s core engine, meaning websites are susceptible even if their specific application code is secure. Both issues have been assigned a critical severity score of 9.2 out of 10.

OpenAI highlighted the incident as a prime example of its “Defender’s Window” concept, a term introduced after its models were used to hack Hugging Face. The company describes this as a narrow timeframe where organizations can leverage advanced AI to find and patch defenses before these capabilities become widely available to malicious actors. OpenAI noted that human developers and security teams had examined the code for two years without fully resolving the issues, whereas AI-assisted researchers developed working exploits in just two weeks.

While some may argue that the flaws were effectively harmless without AI to exploit them, OpenAI challenges this view. The company maintains that powerful technology will inevitably be used by bad actors, creating a collective responsibility to address exploitable flaws proactively. It cited a recent report by Anthropic, whose Claude Mythos model identified thousands of zero-day vulnerabilities in major operating systems and web browsers, as further evidence of the accelerating threat landscape.

The company considers this one of the most significant cybersecurity discoveries of 2026 due to the ubiquity of PHP. By demonstrating the speed at which AI can turn latent bugs into active threats, OpenAI aims to underscore the urgency of the Defender’s Window. The organization continues to collaborate with partners in the Patch the Planet program to identify and mitigate similar risks across the digital infrastructure.