GPT-5.5-Cyber built a zlib fuzzing lab in a day (blog.trailofbits.com)

In a remarkable showcase of AI capability, GPT-5.5-Cyber successfully built a fuzzing lab for the zlib compression library in just one day, a task that traditionally would have taken weeks for skilled researchers. This effort is part of the "Patch the Planet" initiative, a collaboration with OpenAI aimed at addressing the challenges faced by open-source software (OSS) maintainers in managing a surge of security bug reports generated by powerful models. By leveraging the /goal command, GPT-5.5-Cyber autonomously determined that dynamic testing of the zlib code was more effective than static code reviews, leading to the identification of critical vulnerabilities in the library. The significance of this development lies in its potential to revolutionize security testing for OSS projects. As traditional barriers in setting up fuzzing campaigns diminish, both skilled researchers and less experienced attackers will have easier access to potent testing tools. This shift will enhance the speed and efficiency of code reviews, but it also raises concerns about security, as attackers can potentially exploit these same capabilities. With the ability to assess the severity of findings accurately and self-manage the fuzzing process, GPT-5.5-Cyber demonstrates that AI advancements could alter the landscape of software security, emphasizing the need for preemptive measures by developers to address vulnerabilities before they can be exploited.