As concerns over AI-powered hacking and other security risks continue to grow, OpenAI on Monday announced a series of cybersecurity-focused initiatives.
These include an improved version of its limited-access security model, GPT-5.5-Cyber, expanded collaborations with governments and institutions through its “Trusted Access for Cyber” program, and the release of its Codex security scanner as an app plugin.
OpenAI also unveiled Patch the Planet, an initiative launched alongside security research firm Trail of Bits and in collaboration with vulnerability management platforms HackerOne and Calypso.
According to Trail of Bits co-founder and CEO Dan Guido, the project aims to help open-source software communities stay ahead of AI-powered bug-hunting tools while demonstrating the benefits of AI-assisted coding.
A flood of low-quality submissions has made it increasingly difficult to identify and prioritize genuine security flaws.
More from Tech Anthropic was given 90 minutes to restrict foreign access to its AI models after potential jailbreak was identified: Report Trump privately ridiculed Zuckerberg and Bezos after election outreach, book claims To kick off the initiative, Trail of Bits recently organized a five-day opening sprint that brought together 25 engineers to work directly with a range of open-source maintainers.
OpenAI and Trail of Bits said the project uncovered hundreds of bugs and produced dozens of patches in its first week alone.
Guido added that, with funding from OpenAI and unmetered access to its models, Trail of Bits plans to continue investing heavily in the Patch the Planet initiative over the long term.
The announcements come amid intensifying competition in AI cybersecurity.
Earlier this month, rival Anthropic withdrew its Fable 5 and Mythos 5 models following concerns from the Trump administration about their cybersecurity capabilities.
The White House later imposed export restrictions on the models, arguing that safeguards limiting their advanced biological and cyber-related functions were inadequate.
Quick Reads View All 'You can't call it progress': Microsoft CEO Satya Nadella warns against concentration of AI power From iPhone 18 Pro to iPhone Fold: All the iPhones Apple could launch in 2026 Unlike Anthropic’s release, OpenAI’s new GPT-5.5-Cyber checkpoint is not being made publicly available.
Instead, it is being offered through the company’s restricted Trusted Access for Cyber program.
OpenAI said the model scored 85.6 percent on the CyberGym benchmark, outperforming Anthropic’s Mythos 5, which achieved 83.8 percent.
Amid the growing AI cybersecurity race, the Five Eyes intelligence alliance issued a rare joint warning, saying frontier AI models could transform both offensive and defensive cyber operations within months rather than years, underscoring the urgent need for stronger cyber resilience.
