OpenAI launches 'Patch the Planet' and upgrades GPT-5.5-Cyber to take on Anthropic

OpenAI is releasing a more permissive version of GPT-5.5-Cyber, designed for advanced, authorized security work and available only to vetted cybersecurity companies and researchers. According to the company's blog, the updated model can perform deeper analysis across large codebases, identify security-relevant components, validate likely vulnerabilities, and develop and test software patches. It achieved 85.6% on CyberGym — an internal benchmark measuring whether an agent can reproduce known software vulnerabilities — versus 81.8% for GPT-5.5.

The launch is wrapped in two new programs. The OpenAI Daybreak Cyber Partner Program lets participating security vendors use GPT-5.5 with 'Trusted Access for Cyber' inside the products they sell to customers — a shift from the prior model where approved orgs primarily used cyber models on systems they owned or were authorized to test. 'Patch the Planet,' founded with Trail of Bits and developed with HackerOne and Calif, aims to help open-source maintainers manage and remediate vulnerabilities at machine speed. Sam Altman framed it as moving 'to solve security problems instead of just finding them.'

The timing is pointed. The release comes as Anthropic remains in limbo with the U.S. government following the global suspension of its Mythos-class Fable 5 over a jailbreak flaw, and as a Five Eyes joint warning says frontier models will transform offensive and defensive cyber 'in months, not years.' OpenAI's bet is that vetted, gated access threads the needle between empowering defenders and limiting malicious use — but skeptics on Hacker News warn the constant patching treadmill could itself create silent security gaps, and that 'authorized use' gating is hard to police at scale.