PHP Vulnerabilities Boost OpenAI’s Cybersecurity Push

In August, multiple research teams at different organizations used AI models to find two vulnerabilities in PHP, a programming language used by an estimated 70% of the web. The flaws and their exploitation method were grouped under the name “php2shell.” OpenAI’s GPT-5.6 Sol appears to have played a key role in finding them, while Google’s Gemini 4 Argon and other models also seem to have contributed. The PHP Foundation announced on Oct. 20 that it would issue a critical patch within 24 hours. A coalition including OpenAI, Wiz and Calif disclosed the vulnerabilities the following day, after working privately on a fix for two months.

Amazon Web Services said hackers began exploiting php2shell on Oct. 17, before the public disclosure. Several organizations had been notified in advance and agreed to publish information at 14:00 UTC on Oct. 21, but news of the flaws and how to exploit them appears to have leaked beforehand. After the disclosure, websites built on vulnerable PHP code faced “millions” of exploitation attempts as operators worked to install the patch. WordPress reported no issues, while some governments and private organizations took web services offline until they could apply the fix, temporarily preventing people from using those services.

The PHP incident has raised awareness of OpenAI’s Daybreak, an application-only program that gives selected organizations access to its advanced tools for defending against sophisticated AI-powered attacks. OpenAI reported two weeks ago that participants had found 1,885 software flaws, up from 858, and produced 661 patches, up from 263. OpenAI cofounder and president Greg Brockman promoted the program in a lengthy post on X about the exploit and the company’s role in discovering it. Over the summer, OpenAI also hired a chief revenue officer with a cybersecurity background and tasked him with building a sales team for its defensive software.

OpenAI plans to debut a cybersecurity-focused model this fall, along with a tool to help businesses deploy it across their infrastructure. Fortune reports the company is betting that AI cybersecurity tools will become a source of revenue; Anthropic also runs Project Glasswing, a program for cybersecurity professionals at vetted organizations, and other software-as-a-service providers sell AI-powered cyber defense tools. The newsletter also cites experts who say AI models have recently improved at processing large amounts of code and acting autonomously, while hackers increasingly use groups of AI agents, known as agent swarms, to move faster than humans can keep up with.

In August, multiple research teams at different organizations used AI models to find two vulnerabilities in PHP, a programming language used by an estimated 70% of the web. The flaws and their exploitation method were grouped under the name “php2shell.” OpenAI’s GPT-5.6 Sol appears to have played a key role in finding them, while Google’s Gemini 4 Argon and other models also seem to have contributed. The PHP Foundation announced on Oct. 20 that it would issue a critical patch within 24 hours. A coalition including OpenAI, Wiz and Calif disclosed the vulnerabilities the following day, after working privately on a fix for two months.

Amazon Web Services said hackers began exploiting php2shell on Oct. 17, before the public disclosure. Several organizations had been notified in advance and agreed to publish information at 14:00 UTC on Oct. 21, but news of the flaws and how to exploit them appears to have leaked beforehand. After the disclosure, websites built on vulnerable PHP code faced “millions” of exploitation attempts as operators worked to install the patch. WordPress reported no issues, while some governments and private organizations took web services offline until they could apply the fix, temporarily preventing people from using those services.

The PHP incident has raised awareness of OpenAI’s Daybreak, an application-only program that gives selected organizations access to its advanced tools for defending against sophisticated AI-powered attacks. OpenAI reported two weeks ago that participants had found 1,885 software flaws, up from 858, and produced 661 patches, up from 263. OpenAI cofounder and president Greg Brockman promoted the program in a lengthy post on X about the exploit and the company’s role in discovering it. Over the summer, OpenAI also hired a chief revenue officer with a cybersecurity background and tasked him with building a sales team for its defensive software.

OpenAI plans to debut a cybersecurity-focused model this fall, along with a tool to help businesses deploy it across their infrastructure. Fortune reports the company is betting that AI cybersecurity tools will become a source of revenue; Anthropic also runs Project Glasswing, a program for cybersecurity professionals at vetted organizations, and other software-as-a-service providers sell AI-powered cyber defense tools. The newsletter also cites experts who say AI models have recently improved at processing large amounts of code and acting autonomously, while hackers increasingly use groups of AI agents, known as agent swarms, to move faster than humans can keep up with.