70% of the web at risk as AI finds 2-year-old PHP bug

The web infrastructure running the majority of global sites has been shaken by a discovery that highlights the rapid shift in cybersecurity capabilities. On Tuesday, October 21, OpenAI confirmed that its GPT-5.6 Sol model helped researchers identify two critical vulnerabilities in PHP, the open-source language that powers an estimated 70% of the internet.

The incident marks a significant milestone for the Patch the Planet initiative. This program brings together cybersecurity firms, government agencies, and software maintainers to leverage AI for defensive purposes. The discovery underscores a new era where machine intelligence is not just assisting but leading the detection of deep, systemic flaws in foundational software.

The timeline of this discovery stretches back further than the initial report suggests. The PHP Foundation was aware of the underlying code flaw since 2024. However, human developers and security teams struggled to find a practical exploitation method or a viable fix for two years. The breakthrough came on July 30, 2026, when Calif, a founding partner of the initiative, began auditing PHP with OpenAI tools at the Foundation's request. By August 11, researchers had identified exploitable pathways for the vulnerability.

Once the flaws were understood, the remediation process proved complex. The issues required major changes to the PHP engine, extending the fix timeline to two months. The PHP Foundation released patched versions 8.2.35, 8.3.36, 8.4.27, and 8.5.12 to address the problems. Sites running PHP 8.1 or older versions do not receive official fixes. The coordinated release was originally planned for Wednesday, October 22, but was moved up to Tuesday after premature disclosures on October 20 sparked public concern.

The two vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, involve fundamental errors in how PHP manages memory and handles errors. Both received a critical severity score of 9.2 out of 10. The first flaw occurs when PHP runs out of allocated memory during a request, potentially leaving incomplete internal data that the engine later treats as valid. The second involves a fatal error that leaves PHP pointing to memory that has already been freed, allowing subsequent requests to corrupt or access that space.

These weaknesses allow attackers to execute malicious code remotely without needing login credentials. Under specific conditions, this could lead to full control of affected websites. AI researchers also demonstrated that the flaws could expose sensitive data stored in server memory, such as passwords, authentication tokens, and API keys. Because the bugs exist in the core engine, websites are vulnerable even if their own application code is secure.

OpenAI frames this incident as an example of the Defender's Window. This concept describes a narrow timeframe where organizations can use advanced AI to find and patch vulnerabilities before those same capabilities become widely available to malicious actors. The company argues that while powerful technology will inevitably be used by bad actors, it is a collective responsibility to act proactively. This view challenges the notion that AI-only exploitable flaws are harmless, citing recent findings by Anthropic's Claude Mythos in major operating systems and browsers.

The rapid exploitation of these two-year-old bugs by AI-assisted researchers in just two weeks demonstrates the changing pace of security threats. As these patches roll out, the incident serves as a stark reminder that the foundational layers of the web are under constant, increasingly automated scrutiny. For Pacific Northwest businesses and developers relying on PHP, the immediate step is clear: upgrade to the patched versions to ensure their digital assets remain secure against this new class of AI-driven threats.