OpenAI Announces Critical PHP Flaws Fixed via AI-Driven Audit

OpenAI disclosed on Wednesday that its artificial intelligence models have identified and helped resolve significant security weaknesses in PHP, the open-source language that powers an estimated 70 percent of the internet. The announcement, made under the company’s “Patch the Planet” cybersecurity initiative, highlights the rapid evolution of AI-assisted code auditing. Following the discovery, OpenAI coordinated with PHP maintainers, security providers, and government agencies to implement fixes over a two-month period. The company noted that the extended timeline was necessary due to the complexity of the required changes to the PHP engine.

The PHP Foundation has issued patched versions, specifically 8.2.35, 8.3.36, 8.4.27, and 8.5.12, to address the issues. Notably, older versions such as PHP 8.1 and below will not receive official patches. While the coordinated release was initially planned for yesterday, it was accelerated to Wednesday in response to premature disclosures that triggered public concern. OpenAI stated that any site using PHP can now safely upgrade to ensure security.

The vulnerabilities were first identified by researchers from Calif, a California-based cybersecurity firm and founding partner of the Patch the Planet program. At the request of the PHP Foundation, the team began auditing the language on July 30. On August 11, researcher Ryan A. identified an exploitable flaw, a finding immediately corroborated by OpenAI’s models. Shortly thereafter, the security firm Wiz independently discovered a second, related vulnerability. Both issues are tracked as CVE-2026-109618 and CVE-2026-109619.

These flaws stem from errors in how PHP manages memory and handles errors. The first vulnerability occurs when the engine runs out of allocated memory during a request, potentially leaving incomplete data that is later misidentified as valid. The second involves a fatal error that leaves the system pointing to already freed memory, allowing subsequent requests to access or corrupt that space. Under specific conditions, attackers can exploit these weaknesses remotely without login credentials to execute malicious code and potentially seize control of affected servers.

Both vulnerabilities were assigned a critical severity score of 9.2 out of 10. Using AI tools, researchers demonstrated that these flaws could expose sensitive data stored in server memory, including passwords, authentication tokens, and API keys. Crucially, because the issues reside in PHP’s core engine, websites can be compromised even if their specific application code contains no security flaws. This widespread impact has led OpenAI to classify the incident as one of the most significant cybersecurity discoveries of 2026.

The incident illustrates what OpenAI terms the “Defender's Window,” a concept introduced after its models previously hacked Hugging Face. This theory posits a narrow timeframe during which organizations can leverage advanced AI to find and patch defenses before malicious actors gain similar capabilities. OpenAI argues that while human developers examined the code for two years without a full resolution, AI-wielding researchers developed practical exploits within two weeks. The company contends that proactive action is a collective responsibility, noting that powerful technology will inevitably be used by bad actors.

However, the situation presents open questions regarding the nature of AI-driven security. Some argue that flaws which require AI to become exploitable were effectively harmless prior to this development. OpenAI challenges this view, pointing to recent findings by Anthropic’s Claude Mythos, which identified thousands of zero-day vulnerabilities in major operating systems and web browsers. As AI capabilities expand, the balance between defensive and offensive applications continues to shift, placing new demands on the software industry to adapt its security protocols.