OpenAI has rolled out a coordinated cybersecurity push featuring an upgraded GPT-5.5-Cyber model and a new open-source bug-fixing program called Patch the Planet, designed to help maintainers tackle the growing backlog of AI-generated vulnerability reports. The headline benchmark score is 85.6% on CyberGym, narrowly beating Anthropic's Mythos 5 at 83.8%, though the model remains locked behind OpenAI's Trusted Access for Cyber program and is not available to the general public.

What Is Patch the Planet and Why Does It Matter?

Patch the Planet is the broader strategic bet. Launched in partnership with security firm Trail of Bits and vulnerability-management partners HackerOne and Calif, the program pairs open-source project maintainers with free security consulting, custom AI agent tooling, six months of free ChatGPT Pro access, and six months of Codex Security, OpenAI's code-scanning product. More than 30 open-source projects have already joined the initiative.

The problem the program addresses is real and growing. AI models can now generate vulnerability reports at scale, but many of these are low-quality or redundant, creating what the industry calls "slop CVEs." Open-source maintainers, who typically work with minimal budgets and no dedicated security staff, have been buried under the triage burden. Trail of Bits CEO Dan Guido emphasized that maintainers "do their work out of love of open source, and now they're stuck reviewing slop CVEs".

Dan Guido

How Does OpenAI Plan to Fix the Open-Source Security Problem?

• Direct Engineering Support: In a five-day opening sprint, Trail of Bits deployed 25 engineers, roughly one-fifth of its entire workforce, on simultaneous collaborations with participating open-source maintainers.
• Subsidized Compute Power: OpenAI has subsidized Codex Security scanner usage to the tune of 20 trillion tokens since the product entered research preview, signaling the company's willingness to absorb significant compute costs to seed adoption.
• Customized Tooling: Rather than applying a one-size-fits-all approach, roughly half of the sprint time went to customizing AI agents to each codebase, ensuring maintainers inherit working tools tailored to their specific projects.
• Measurable Results: In the first week alone, the program produced hundreds of bug discoveries and dozens of landed patches across participating projects.

Fouad Matin, OpenAI's cyber tech lead, explained the economic logic: "The goal is to flip that economics" by having the same class of models generating the noise also clear the backlog, provided someone covers the token bill and engineering hours.

Fouad Matin, OpenAI's cyber tech lead

Why Is This Announcement Strategically Significant?

The timing is not accidental. Earlier this month, Anthropic was forced to pull its Fable 5 and Mythos 5 models off the market after the Trump administration imposed export controls, citing concerns that Anthropic's voluntary safeguards on advanced biological and cybersecurity capabilities were insufficient. That regulatory pressure has created an opening for OpenAI to position itself as the vendor that can ship cyber-capable models under government-acceptable guardrails.

On the same day as OpenAI's announcement, the Five Eyes intelligence alliance, a partnership between the United States, United Kingdom, Canada, Australia, and New Zealand, issued an unusual joint statement on AI cyber risk. The statement compressed the threat timeline from years to months, warning that frontier AI models are anticipated to exceed current industry expectations and fundamentally transform both offensive and defensive cyber capabilities.

"Frontier AI models are anticipated to exceed current industry expectations, fundamentally transforming both offensive and defensive cyber capabilities. The timeline is not years, it is months. In this environment, cyber resilience is integral," stated the Five Eyes intelligence alliance in a joint statement.

Five Eyes intelligence alliance, joint statement

What Are the Limitations and Open Questions?

The benchmark gap between GPT-5.5-Cyber and Mythos 5 is narrow. A 1.8-point difference on a single benchmark does not necessarily reflect the full surface of offensive or defensive work a cybersecurity analyst performs. Additionally, GPT-5.5-Cyber is locked behind Trusted Access, meaning independent researchers cannot replicate the score or stress-test the model against novel codebases.

The open-source community also has legitimate reasons to be cautious. Free tooling that creates long-term dependence on a single vendor's subsidies can become problematic if those subsidies end or if the vendor's priorities shift. Trail of Bits has committed to continuing the work long-term on OpenAI funding plus unmetered model access, but the sustainability of that arrangement remains uncertain.

The harder test arrives over the next two quarters. Whether 20 trillion subsidized tokens actually shrink the maintainer backlog, or whether the slop simply gets faster, will determine whether Patch the Planet becomes a genuine public good or a clever marketing vehicle for OpenAI's cybersecurity capabilities. For now, the program represents a concrete attempt to address a real problem in open-source security, backed by significant compute resources and engineering talent.