OpenAI's new model finds and patches security bugs on its own
OpenAI's new model finds and patches security bugs on its own. OpenAI released GPT-5.5-Cyber, a specialised model built for the full defensive-security loop — finding vulnerabilities, proving they're real, writing the patch, and testing the fix — and launched 'Patch the Planet', where 30-plus open-source projects, including curl, Go, Python and pyca/cryptography, have signed up to receive fixes.
What it actually means
Vulnerability-hunting AI has existed for a while; what's new is a model doing the whole loop without a human stitching the steps together. If you maintain code other people depend on, that's a real offer, not a benchmark headline. Access to the top tier is still gated and vetted, so this isn't 'plug it into your repo today' — but the direction is clear: as agents write more of your code, the tools to check it are catching up. Worth checking whether any of your dependencies are on the Patch the Planet list.
As agents write more of the code, the tools to check it are finally catching up.
30+open-source projects enrolled in OpenAI's Patch the Planet fix programme
Source: OpenAI · Wednesday, 1 July 2026