SAN FRANCISCO — The name comes from a line in a 1995 movie: “Hack the Planet,” shouted by a teenage hacker in the film of the same name. OpenAI’s version is more earnest. The company announced a program called Patch the Planet on Monday, an effort to find and fix security vulnerabilities in open-source software, the shared code that runs a large part of the internet and, increasingly, the AI industry itself.
The program pairs OpenAI with Trail of Bits, a security firm known for auditing code that other companies avoid, and it will put security engineers to work on the open-source projects that underpin modern computing. OpenAI’s own Codex tools, which can analyze and modify code, will assist in the work, according to the company’s announcement. The company is also releasing a new model, GPT-5.5-Cyber, aimed at security tasks, along with a bug bounty program tied to the initiative.
WIRED described the launch as a direct response to Anthropic’s Mythos, a security framework the rival lab published this year that set out its own approach to protecting AI systems and the software around them. The framing matters: OpenAI and Anthropic have spent the past year competing on model quality, enterprise deals and public perception, and security has become the newest front in that rivalry. Each company wants to define the standards for safe AI, and each is trying to make its approach the one governments and customers adopt.
The practical stakes are easier to measure. Open-source software is the load-bearing wall of the digital economy, and its security has been chronically underfunded. The maintainers of widely used libraries are often volunteers, and the vulnerabilities that matter most are discovered by accident or by attackers rather than by defenders. The log4j incident in 2021, which exposed millions of servers to remote attack through a logging library, demonstrated what happens when critical code is maintained by a handful of people with limited resources.
AI has changed the calculus. Research has shown that models can find vulnerabilities in code faster than traditional scanning tools, and security companies have begun integrating AI into their products. The same capability that helps defenders also helps attackers, and the industry has been debating whether AI-assisted vulnerability discovery increases net security or net risk. OpenAI’s program is a bet on the former, on the condition that the discoveries are shared and fixed rather than exploited.
The mechanics of the program matter for its credibility. Trail of Bits engineers will work directly with open-source maintainers, according to the announcement, rather than publishing findings without coordination, an approach designed to avoid the disclosure battles that have plagued security research. The bug bounty program gives financial incentives for independent researchers to participate, and the GPT-5.5-Cyber model is intended to make AI-assisted analysis available to defenders who cannot afford dedicated security teams.
The competitive subtext is unmistakable. Anthropic has positioned itself as the safety-conscious lab, and its Mythos framework was an attempt to set the terms of the security debate. OpenAI’s countermove is to demonstrate that it can secure not only its own systems but the broader software ecosystem, a claim that carries weight with enterprise customers weighing which lab to trust with their infrastructure.
The dual-use problem remains unresolved. The tools that Patch the Planet will use to find vulnerabilities can also be used to exploit them, and the models are being released publicly. OpenAI has said it will publish findings only after fixes are available, and that the security model includes safeguards against misuse, but the fundamental tension is the same one that has defined AI security since the technology became capable: the knowledge that makes systems safer also makes them more vulnerable.
Governments are watching the rivalry with interest. Both labs have courted defense and intelligence agencies, and the ability to secure open-source infrastructure has become a selling point in those conversations. A program that visibly improves the security of widely used code is worth more than any white paper, analysts said, and both companies appear to understand that the standards they establish now will shape the market for years.
The program joins a crowded field of efforts to fund open-source security, and its distinctiveness is the use of AI models in the work itself. The Open Source Security Foundation and the U.S. Cybersecurity and Infrastructure Security Agency have both run programs that direct money and engineers to critical projects, and their results have been uneven, in part because the number of projects dwarfs the resources available. OpenAI’s contribution is different in kind: the same models that the industry worries about as an offensive threat are being pointed at the defensive problem, and the company is betting that its Codex tools can multiply the productivity of the security engineers who use them. If the approach works, it could change the economics of open-source security, which have never favored the defenders.
For the maintainers of open-source projects, the program is a change of fortune. The people who have kept critical libraries alive on donations and evenings now have the attention of the most valuable AI companies in the world, and the promise of professional security review is one they have sought for years. Whether the attention translates into lasting support, or fades when the rivalry moves to the next front, will determine whether Patch the Planet is remembered as a genuine contribution or a chapter in a corporate competition.
