Artificial intelligence models from OpenAI and Google have helped cybersecurity researchers uncover critical vulnerabilities in PHP, a programming language used by a large share of the internet. The vulnerabilities, tracked as CVE-2026-109618 and CVE-2026-109619, could allow attackers to execute malicious code on affected servers without authentication or access sensitive information stored in server memory. Both flaws received a critical severity score of 9.2 out of 10. PHP maintainers released security fixes on October 21, 2026, urging website operators to update immediately.
The discovery involved OpenAI’s GPT-5.6 Sol, Google’s Gemini 4 Argon and security research tools developed by Calif and Wiz. The flaws had remained in PHP’s underlying code for years, despite the language’s widespread use. The coordinated response brought together software maintainers, AI developers, cybersecurity companies and infrastructure providers to prepare patches and protective measures before public disclosure. The incident demonstrates how AI can accelerate vulnerability discovery while also raising concerns that similar tools could help attackers develop exploits more quickly.
Key Takeaways
- Two critical PHP vulnerabilities were disclosed: CVE-2026-109618 and CVE-2026-109619 affect PHP’s core runtime.
- AI played a central role in discovery: GPT-5.6 Sol helped Calif identify a flaw, while Wiz independently used Gemini 4 Argon, GPT-5.6 Sol and its Wiz Atlas platform.
- The potential impact is extensive: PHP supports websites and applications built with WordPress, Drupal, Joomla and other widely used software.
- Both vulnerabilities scored 9.2 out of 10: Successful exploitation can enable remote code execution or expose sensitive process memory.
- Patches are available: Fixed releases include PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35.
- Older PHP versions need special attention: PHP 8.1 and earlier are no longer supported and do not receive fixes for these vulnerabilities.
What Are the PHP Vulnerabilities?
PHP is a server-side programming language that helps websites generate dynamic pages, process requests and interact with databases. It is widely used by content management systems, online stores, publishing platforms and web applications.
The newly disclosed vulnerabilities affect PHP’s core engine rather than a particular website plugin or theme. That distinction is important because even an application with well-written code may remain exposed if it runs on a vulnerable PHP version.
The flaws involve memory-safety problems that can occur when PHP handles certain fatal errors and request-processing conditions. Under particular circumstances, internal state can become corrupted or remain accessible after the memory it refers to has been released.
Researchers demonstrated that these conditions could lead to serious security consequences, including remote code execution and disclosure of sensitive memory.
| Vulnerability | Potential impact | Severity |
|---|---|---|
| CVE-2026-109618 | Remote code execution and process-memory disclosure | Critical, 9.2/10 |
| CVE-2026-109619 | Remote code execution linked to stale internal state | Critical, 9.2/10 |
The two vulnerabilities are collectively referred to in research coverage as PHP2Shell. Their technical details differ, but both expose weaknesses in how PHP manages internal memory and request state.
How GPT-5.6 Sol and Gemini 4 Argon Helped Find the Flaws
The discovery was not the work of a single AI model or company. Multiple research teams investigated PHP independently and contributed to a coordinated disclosure process.
OpenAI and Calif used GPT-5.6 Sol
According to OpenAI, the cybersecurity company Calif identified a critical flaw on August 1, 2026, with the assistance of GPT-5.6 Sol as part of OpenAI’s Patch the Planet initiative.
The initiative focuses on finding vulnerabilities in widely used software and coordinating with maintainers to fix them. OpenAI subsequently investigated the issue and worked with the PHP maintainers and other security partners.
Wiz combined Gemini 4 Argon with GPT-5.6 Sol
Wiz Research independently discovered the vulnerabilities using its Wiz Atlas research platform alongside Google’s Gemini 4 Argon and OpenAI’s GPT-5.6 Sol.
Google Cloud said Gemini 4 Argon contributed to vulnerability discovery, exploit development, attack detection and patch validation. Wiz also described how AI helped researchers adapt and test exploit techniques across different software configurations.
These capabilities matter because a vulnerability may be difficult to exploit even after a weakness has been identified. Researchers often need to understand how a program behaves under unusual conditions, develop a reliable demonstration and verify that a proposed patch actually closes the security gap.
AI tools helped reduce the time and complexity involved in that work.
Why Millions of Websites Could Be at Risk
PHP remains one of the most widely used technologies for running websites. It underpins popular platforms such as WordPress, Drupal and Joomla, as well as numerous custom applications and online services.
Wiz estimates that PHP powers approximately 70% of websites worldwide. That figure describes PHP’s broad reach, not the percentage of websites confirmed to be vulnerable.
Reuters reported that data from internet intelligence company Censys identified approximately 1.5 million of 2.6 million hosts reporting a PHP version for which no fix was available at the time of its report. These figures indicate a substantial potential exposure, but they should not be interpreted as a verified count of successfully compromised websites.
The vulnerabilities affect PHP versions dating back to 5.4, according to Wiz. However, the actual risk to any particular website depends on its runtime version, deployment configuration, available mitigations and whether the system has been patched.
The incident also illustrates why vulnerabilities in shared infrastructure can have consequences far beyond one application. A flaw in a widely used runtime can affect many unrelated websites simultaneously.
PHP Releases That Fix the Vulnerabilities
PHP maintainers released patched versions on October 21, 2026. Organizations should verify their installed version and apply the appropriate update.
| PHP branch | Patched release |
|---|---|
| PHP 8.5 | 8.5.12 |
| PHP 8.4 | 8.4.27 |
| PHP 8.3 | 8.3.36 |
| PHP 8.2 | 8.2.35 |
These are the specific fixed releases identified in the security guidance available at publication. Administrators should consult the official PHP advisory for the latest maintenance releases and installation instructions.
PHP 8.1 and earlier are end-of-life versions and do not receive fixes for these vulnerabilities. Organizations still using them should plan a migration to a supported PHP branch rather than assume that a routine update will resolve the problem.
What website owners should do
Website owners and administrators should:
- Check the PHP version: Determine which runtime powers each website, application, container or hosting environment.
- Install a patched release: Upgrade to a fixed supported version and redeploy affected services where necessary.
- Review hosting responsibilities: Managed hosting customers should confirm whether their provider has applied the update. Self-managed servers and containers may require manual action.
- Check for suspicious activity: Review server logs, security alerts and access records for unusual behaviour, especially if the system remained exposed after disclosure.
- Use temporary protections if patching is delayed: Follow guidance from the PHP project, hosting provider and security teams. Taking an exposed service offline may be necessary when effective mitigation is unavailable.
A version check alone is not a complete security review. Organizations should also confirm that the running service actually uses the patched runtime and that old containers or deployment images have not been left in production.
The Growing Role of AI in Cybersecurity
The PHP incident highlights a change in how security research can be conducted. AI models can help researchers inspect complex code, generate hypotheses about possible failures, test edge cases and analyse the results of experiments.
These capabilities can help defenders uncover weaknesses that have survived years of conventional review. In this case, the vulnerabilities were found in a mature and widely scrutinized programming language.
However, the same capabilities can create risks. Once a vulnerability becomes public and a patch is available, attackers may use AI to understand the weakness, adapt exploit attempts and identify systems that remain unpatched.
Reuters reported that security firms had observed early attempts to exploit the vulnerabilities around the disclosure period. The extent of successful exploitation was still developing, underscoring the importance of distinguishing attempted attacks from confirmed compromises.
The response also shows why vulnerability discovery alone is not enough. Researchers must coordinate with maintainers, test fixes, inform service providers and give organizations time to protect their systems.
The Bigger Picture
PHP2Shell illustrates the dual-use nature of advanced AI in cybersecurity. The same models that can help developers find and fix defects can potentially help malicious actors exploit them. As AI-assisted research becomes more capable, the time between discovering a vulnerability and seeing exploitation attempts could shrink.
The case also reinforces the importance of maintaining widely used software. A single weakness in a core runtime can create risk across thousands of unrelated applications, regardless of how carefully individual sites have been built.
For businesses, the immediate priority is to patch vulnerable systems, verify deployments and review potential signs of compromise. For the cybersecurity industry, the longer-term challenge is to make AI-assisted vulnerability discovery part of a repeatable process that consistently turns findings into protection.
FAQs
1. What are the PHP2Shell vulnerabilities?
PHP2Shell refers to two critical PHP vulnerabilities, CVE-2026-109618 and CVE-2026-109619. They can potentially allow remote code execution or disclosure of sensitive process memory on affected systems.
2. Which AI models helped discover the PHP flaws?
OpenAI’s GPT-5.6 Sol helped Calif identify a critical flaw. Wiz independently used Gemini 4 Argon, GPT-5.6 Sol and its Wiz Atlas platform during vulnerability research and remediation.
3. Which PHP versions are affected?
The vulnerabilities affect PHP versions dating back to 5.4. Fixed releases include PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. PHP 8.1 and earlier are end-of-life and require migration to a supported version.
4. Are all PHP websites compromised?
No. The vulnerability creates potential exposure, but that does not mean every affected website has been attacked or compromised. Risk depends on the runtime version, configuration, mitigations and patch status.
5. What should WordPress and Drupal website owners do?
They should check which PHP version their hosting environment uses, confirm whether the host has applied a fixed release, and contact the provider or update the runtime if necessary. They should also review security logs if the website was exposed.
Looking Ahead
The coming weeks will show how quickly website operators can apply the fixes and whether attackers can exploit vulnerable systems at scale. Hosting companies, software vendors and security teams will need to coordinate closely, particularly for older deployments that cannot be patched without upgrading their PHP version or changing their infrastructure.
The wider lesson is that AI is changing both vulnerability research and the potential speed of exploitation. The most effective response will combine AI-assisted detection with secure software maintenance, rapid patching, clear disclosure procedures and continuous monitoring of systems that remain exposed.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
