70% of the web targeted as AI uncovers two critical PHP flaws
OpenAI’s GPT-5.6 Sol model helped researchers turn a dormant code issue into a working exploit, prompting an emergency patch for the language that powers most of the internet.
By Rafael Collins ·
The internet’s backbone faces a rare and urgent threat, with researchers confirming that two critical security flaws in PHP, the open-source language powering an estimated 70% of the web, have been weaponized. The discovery, announced on Oct. 21 by OpenAI, highlights a widening gap between the speed of AI-assisted hacking and the pace of traditional human-led security development. The vulnerabilities, which had stumped engineers for two years, were turned into working exploits in just two weeks, forcing a coordinated, emergency response across the global tech sector.
The incident was uncovered through OpenAI’s “Patch the Planet” cybersecurity initiative, which utilized its GPT-5.6 Sol model to audit the PHP codebase. The process began on July 30, 2026, at the request of the PHP Foundation. On Aug. 11, a researcher at Calif, a California-based cybersecurity firm and founding partner of the program, identified an exploitable weakness. OpenAI researchers confirmed the finding the same day, while security firm Wiz independently discovered a second, related vulnerability. Together, these flaws allow attackers to remotely execute malicious code on servers without needing login credentials, potentially granting them full control of affected websites.
The two issues, designated as CVE-2026-109618 and CVE-2026-109619, both stem from how PHP manages computer memory and processes errors. The first occurs when the engine runs out of allocated memory, leaving behind incomplete data that may later be mistakenly treated as valid. The second involves a fatal error that leaves the system pointing to memory that has already been freed, allowing subsequent requests to access or corrupt that space. Both vulnerabilities carry a critical severity score of 9.2 out of 10, and because they reside in PHP’s core engine, they can affect sites even if the application code itself is secure.
In addition to remote code execution, the flaws could expose sensitive data stored in server memory, including passwords, authentication tokens, and API keys. The rapid progression from a theoretical flaw to a practical exploit has drawn attention to the concept of the “Defender's Window.” OpenAI, which introduced this framework after its models previously hacked Hugging Face, describes it as a narrow timeframe during which organizations can use advanced AI to find and patch defenses before those same capabilities become widely available to malicious actors. The company views this PHP incident as a prime example of that window being utilized effectively.
The PHP Foundation has released patched versions, 8.2.35, 8.3.36, 8.4.27, and 8.5.12, to address the issues. Older versions, specifically PHP 8.1 and below, will not receive official fixes, meaning users of those builds must upgrade to newer releases to maintain security. Although the coordinated release was originally scheduled for Oct. 22, it was brought forward to Oct. 21 after premature disclosures on Oct. 20 sparked public outcry. The company states that every site using PHP can now safely upgrade to its latest version to remain protected.
The timeline for the fix was extensive, taking over two months to coordinate a coalition of researchers, PHP maintainers, security providers, and government cybersecurity agencies. OpenAI attributed this duration to the complexity of the task, which required major changes to the PHP engine itself. The underlying flaw had been known to the Foundation since 2024, but human developers were unable to identify a method of exploitation or a suitable fix. It was only the application of AI tools that provided the practical means to both exploit and ultimately resolve the issue.
OpenAI has framed the event as one of the most significant cybersecurity discoveries of 2026, given the ubiquity of the affected language. The company challenges the notion that these flaws were harmless because they would not have been exploitable without AI. Instead, it argues that powerful technology will inevitably be used by bad actors, creating a collective responsibility to act proactively. This perspective is echoed by recent findings from Anthropic, whose Claude Mythos model identified thousands of zero-day vulnerabilities across major operating systems and web browsers.
The incident underscores a shifting dynamic in cybersecurity where AI is no longer just a tool for developers but a primary driver of both vulnerability discovery and mitigation. For the millions of websites running on PHP, the immediate next step is a critical upgrade to ensure the two newly patched flaws do not become a widespread breach. As the “Defender's Window” narrows, the industry is increasingly relying on AI to keep pace with the rapidly evolving threat landscape.
