70 Percent of the Web at Risk: AI Finds Fatal Flaw in PHP Core

Seventy percent of the internet runs on PHP, a statistic that suddenly became a source of concern this week. OpenAI revealed on October 21 that its GPT-5.6 Sol model, working under the company's Patch the Planet cybersecurity initiative, uncovered critical vulnerabilities in the language's core engine. These flaws could allow remote attackers to seize control of servers without needing login credentials. The discovery highlights a significant shift in cybersecurity, where advanced artificial intelligence is now identifying weaknesses that have eluded human developers for years.

The timeline of this crisis began in 2024, when the PHP Foundation first noticed an underlying flaw in the system. For two years, human researchers examined the code but failed to identify a method to exploit the bug or a viable fix. It was not until July 30, 2026, that the PHP Foundation requested assistance from Calif, a California-based cybersecurity firm and founding partner of the Patch the Planet program. Researchers from Calif began auditing the software using OpenAI tools on that date.

On August 11, 2026, Calif researcher Ryan A. identified the first exploitable vulnerability. OpenAI researchers confirmed the finding the same day. Shortly after, another security firm, Wiz, independently discovered a second, related vulnerability. These two issues are tracked as CVE-2026-109618 and CVE-2026-109619. Both stem from how PHP manages computer memory and processes errors, creating a narrow window of opportunity for attackers to execute malicious code.

The specific mechanics of the flaws are technical but severe. The first vulnerability occurs when PHP runs out of allocated memory during a request. This can leave incomplete internal data that the engine later mistakes for valid information. The second flaw involves a fatal error leaving the system pointing to memory that has already been freed. Subsequent requests can then access or corrupt that memory. Under certain conditions, these weaknesses allow for remote exploitation, potentially exposing sensitive data such as passwords, authentication tokens, and API keys stored in server memory.

The coordinated release of fixed versions was originally planned for October 22. However, premature disclosures on October 20 led to public outcry, prompting the PHP Foundation to bring the release forward to October 21. The foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the issues. Versions 8.1 and older will not receive official fixes, meaning site administrators using those iterations must upgrade to stay secure. OpenAI noted that every site using the language can now safely upgrade.

OpenAI frames this incident as a prime example of its Defender's Window concept. The company introduced this idea after its models hacked the platform Hugging Face. The concept describes a critical, narrow timeframe during which organizations can use frontier AI models to find and patch defenses before those same capabilities become widely available to malicious actors. OpenAI argues that it is a collective responsibility to act proactively, given that powerful technology will inevitably be used by bad actors.

The severity of the situation is underscored by the fact that both vulnerabilities received a critical severity score of 9.2 out of 10. The flaws reside in PHP's core engine, meaning websites are affected even if their application code is secure. While human teams spent two years unable to resolve the issues, AI-wielding researchers demonstrated practical exploits within two weeks. This speed raises questions about the future of software security, particularly as other AI systems, such as Anthropic's Claude Mythos, continue to identify thousands of zero-day vulnerabilities across major operating systems and browsers.

Looking ahead, the industry is now focused on rapid adoption of the new patches. The incident serves as a warning that the landscape of digital security is changing rapidly. As AI models become more adept at finding exploits, the window for defenders to secure their systems shrinks. Organizations must now consider leveraging similar AI tools to stay ahead of threats that may be invisible to traditional human auditing methods.