OpenAI AI-Discovered PHP Vulnerabilities: What Site Operators Need to Know
OpenAI says its Patch the Planet initiative helped researchers identify critical vulnerabilities in PHP, the programming language used by many websites and web applications. Researchers at security firm Calif used OpenAI’s GPT-5.6 Sol during an audit, while Wiz independently discovered and reported both vulnerabilities. PHP’s maintainers coordinated the fixes and released security updates on October 21, 2026.
The news is a notable example of AI-assisted security research, but it is also a practical reminder: organizations that run PHP should confirm which software version serves their websites and apply the update recommended by PHP or their hosting and software vendor. These are vulnerabilities in PHP itself—not a finding that every PHP-based site has been compromised. OpenAI’s disclosure describes the research and response, while PHP’s own advisory gives operators the authoritative update instructions. ([openai.com](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/))
What did OpenAI and its partners find?
The disclosure concerns two critical PHP vulnerabilities, CVE-2026-109618 and CVE-2026-109619. PHP’s security advisory says the flaws can allow unauthenticated remote code execution under the right conditions. In plain language, a remote attacker may be able to send a request without first logging in and cause vulnerable PHP software to run code on the server.
The PHP project assigned each vulnerability a CVSS v4.0 severity score of 9.2 out of 10. The score signals a serious risk, but it does not mean every server is exposed in the same way, or that a particular website has been attacked. PHP’s advisory identifies web-serving installations—including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed—as potentially at risk. The project calls out mod_php as a particularly concerning setup because a demonstrated attack could be triggered before application code runs. ([php.net](https://www.php.net/?=PHPE9568F35-D428-11d2-A769-00AA001ACF42))
PHP powers a large share of the web. Tom’s Hardware cited an estimate of around 70% of websites, a figure that conveys the possible scale but should not be read as a count of vulnerable or compromised sites. Exposure depends on the PHP version actually running, the deployment and whether the relevant fix has been applied. Tom’s Hardware’s coverage summarizes the announcement and its potential reach. ([tomshardware.com](https://www.tomshardware.com/tech-industry/cyber-security/openai-says-gpt-5-6-sol-helped-uncover-critical-php-flaws-that-could-affect-70-percent-of-websites-ai-assisted-researchers-turned-two-year-old-bug-that-stumped-human-developers-into-working-exploits-in-two-weeks))
How AI-assisted research contributed
OpenAI says Calif identified a critical flaw using GPT-5.6 Sol as part of Patch the Planet. The PHP Foundation’s account adds important context: Calif reported an unconditional remote-code-execution issue in August 2026, OpenAI reported the same underlying cause independently, and Wiz later contributed additional research. The reports helped clarify the security impact of a long-running problem involving PHP’s out-of-memory handling. ([openai.com](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/))
The Foundation says the issue had previously been difficult to resolve. Developers had considered the out-of-memory problem as far back as 2024, but a workable path to remote code execution was not then established. Possible fixes also raised compatibility and denial-of-service concerns. Later research gave maintainers a clearer understanding of the attack paths and urgency. That is a more precise account than saying an AI simply found a fully understood bug that humans had overlooked: AI-assisted work formed part of a broader process of investigation, verification and engineering.
OpenAI describes Patch the Planet as a coordinated defensive effort involving software maintainers, security researchers, infrastructure providers and other partners. In this case, PHP maintainers led the fixes and release. OpenAI, Calif, Wiz, Cloudflare and others contributed research, testing, technical support or coordination. OpenAI says the process included preparation for disclosure and testing fixes against potential variants. ([openai.com](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/))
That human role matters. A model can help researchers examine code, develop or test hypotheses and surface possible attack paths. It does not, by itself, determine whether a finding is real, select a safe fix for a widely used runtime, review compatibility trade-offs or deploy an update to production. The PHP Foundation reports that people reviewed reports and made decisions about the patches; it also notes that AI-assisted research has increased the volume of reports its security team must triage. ([thephp.foundation](https://thephp.foundation/blog/2026/10/21/todays-php-security-release-and-how-were-strengthening-php-security/))
Which PHP versions received fixes?
On October 21, PHP announced patched releases for four maintained branches:
- PHP 8.5.12
- PHP 8.4.27
- PHP 8.3.36
- PHP 8.2.35
These releases address CVE-2026-109618 and CVE-2026-109619. The version numbers above are the upstream PHP releases announced for those branches; they are not a substitute for checking your operating system, hosting provider or vendor guidance. Distributions and long-term-support vendors may deliver fixes as their own packages, sometimes without changing the upstream-looking version string in the way an operator expects. PHP’s advisory says to check with the vendor for updated packages. ([php.net](https://www.php.net/?=PHPE9568F35-D428-11d2-A769-00AA001ACF42))
PHP 8.1 and older no longer receive updates from the PHP project. The PHP team says some distributions and LTS vendors may provide backported fixes, and it published minimal source backports for PHP 7.4, 8.0 and 8.1 for users building from source or without access to an LTS vendor. If your installation runs an older branch, ask your provider specifically whether it has delivered a fix for these CVEs. If it has not, plan a move to a supported, patched runtime. Do not assume that an old version is safe simply because a hosting dashboard reports that the server is “managed.” ([php.net](https://www.php.net/?=PHPE9568F35-D428-11d2-A769-00AA001ACF42))
Practical checklist for website owners and administrators
- Find out which PHP runtime serves the site. Check your hosting control panel, application platform or server configuration. A command such as
php -vmay show the command-line version, which can differ from the version used by the website. If you do not administer the server, contact your host and ask which PHP version serves your site and whether it is patched for CVE-2026-109618 and CVE-2026-109619. - Use the update for your specific environment. For a directly managed upstream installation on a listed branch, compare against the fixed release for that branch. For packages from a Linux distribution, managed host or commercial LTS provider, follow that vendor’s advisory and package instructions instead. PHP’s official security guidance and downloads are the best starting points; vendors may have their own patch identifiers and procedures. ([php.net](https://www.php.net/?=PHPE9568F35-D428-11d2-A769-00AA001ACF42))
- Apply the update promptly, then verify it is active. Follow the provider’s instructions for updating the PHP runtime and restarting or reloading the relevant service where required. Confirm the version or vendor patch status for the web-serving process—not only for a local command-line installation. Retest important site functions after the change.
- Check every relevant environment. Businesses may have separate PHP versions for production, staging, background jobs, containers or older applications. Inventory internet-facing systems first, then check other installations that process untrusted requests or share production data.
- Treat temporary protections as a bridge, not a replacement for patching. A web application firewall or hosting-provider rule may reduce exposure in some situations, but PHP says there are no complete mitigations for these vulnerabilities: updating is the safe course. Ask your provider what temporary controls are in place while the official fix is being deployed. ([php.net](https://www.php.net/?=PHPE9568F35-D428-11d2-A769-00AA001ACF42))
- Keep a record of the change. Note the affected systems, the package or runtime updated, when the change was made and how you verified it. This makes it easier to answer security, compliance and incident-response questions later.
What this means for WordPress sites
WordPress runs on PHP, so WordPress site owners should make sure their hosting environment is covered by the PHP update. The finding concerns the PHP runtime; it is not, by itself, evidence of a flaw in WordPress core, a specific plugin or a particular theme. Updating WordPress plugins alone will not patch the PHP engine underneath the site.
Many WordPress users cannot install PHP server updates directly. In that case, ask the hosting provider for written confirmation that the PHP version serving the site has the relevant fixes, or that the provider has applied an equivalent vendor backport. If you manage your own server, use the official PHP and operating-system vendor instructions. After the update, check that the site loads, that key forms and administrative functions work, and that scheduled jobs or integrations still run.
What the disclosure does—and does not—show
The incident demonstrates the defensive potential of frontier AI models: AI-assisted research can help experts find or understand flaws in widely deployed software, potentially giving maintainers time to prepare a fix before public details are available. OpenAI calls this a “Defender’s Window.” In the PHP case, the research was followed by coordinated work among maintainers and security partners to prepare updates and guidance. ([openai.com](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/))
It also highlights the dual-use risk. The same kinds of capabilities that help defenders analyze a vulnerability may help attackers understand it or develop offensive tools. OpenAI itself says advanced models can support rapid exploit development. That is why responsible disclosure, careful coordination and timely patching remain essential even when a finding originates in defensive research. ([openai.com](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/))
Finally, a vulnerability disclosure is not proof that every affected system was exploited—or that exploitation was impossible before disclosure. The right response is to establish what software is running, apply the relevant vendor fix and investigate further if logs or other evidence indicate suspicious activity. Avoid both extremes: do not treat every PHP website as confirmed compromised, and do not dismiss a critical runtime issue because your site appears to work normally.
Frequently asked questions
Were these vulnerabilities found by AI alone?
No. OpenAI says GPT-5.6 Sol helped Calif identify a critical flaw, and the PHP Foundation describes contributions from Calif, OpenAI and Wiz. PHP maintainers reviewed and fixed the issues, with people making decisions about the reports and patches. “AI-assisted” is more accurate than saying AI independently handled discovery through remediation. ([openai.com](https://openai.com/index/coordinated-disclosure-of-critical-vulnerabilities/))
Is my PHP or WordPress site already compromised?
The disclosure does not establish that your specific site was attacked. It identifies vulnerabilities that can permit remote code execution in affected PHP installations. Check your runtime and vendor patch status, apply the fix and investigate suspicious activity if you have concrete indicators of a compromise.
Can I rely on a firewall instead of updating PHP?
No. A firewall or other provider-side control may be useful as an interim risk-reduction measure, but PHP says there are no complete mitigations and recommends updating. Confirm the patch status with your host or vendor. ([php.net](https://www.php.net/?=PHPE9568F35-D428-11d2-A769-00AA001ACF42))
What if I cannot find one of the listed fixed version numbers?
Your platform may distribute a vendor-maintained package or backport rather than the exact upstream release number. Check the vendor’s security advisory or ask its support team whether the fix for both CVEs is included. If the installation is on an unsupported branch and no fix is available, prioritize migration to a supported patched version.
Bottom line
The OpenAI AI-discovered PHP vulnerabilities story is both a cybersecurity milestone and an operational reminder. GPT-5.6 Sol contributed to research under Patch the Planet, but maintainers and human security professionals validated the findings, coordinated the response and shipped the fixes. Site owners should focus on the next step: identify the PHP runtime behind each website, consult official PHP or provider guidance, install the applicable update and verify that the patched runtime is serving requests.
One Comment on “OpenAI AI-Discovered PHP Vulnerabilities: What Site Operators Need to Know”