OpenAI Wants Its AI to Patch the World's Software
OpenAI is pointing its newest model at the internet's broken locks. It expanded Daybreak, its cybersecurity platform, with the full release of GPT-5.5-Cyber, an AI built to find and fix software flaws, plus a program called Patch the Planet aimed at securing the open-source code that runs almost everything.
OpenAI is pointing its newest model at the internet's broken locks. It expanded Daybreak, its cybersecurity platform, with the full release of GPT-5.5-Cyber, an AI built to find and fix software flaws, plus a program called Patch the Planet aimed at securing the open-source code that runs almost everything. The idea is defense at machine speed.
Here is how it works. Daybreak pairs GPT-5.5-Cyber with a Codex Security plugin so the AI can scan code, confirm a vulnerability is real, and write the patch, the full loop that used to take human researchers days. OpenAI also launched a Cyber Partner Program with more than 20 security vendors. The pitch is that defenders finally get a tool that scales the way attackers already do.
The benchmark is the headline. GPT-5.5-Cyber scored 85.6 percent on CyberGym, a test of real-world vulnerability finding, the highest any single model has posted. Standard GPT-5.5 scored 81.8 percent, and Anthropic's Claude Opus 4.7 scored 73.1 percent. The model is purpose-built for security work, and the gap over general-purpose systems is the whole selling point.
Patch the Planet aims at the soft underbelly of software. Open-source code sits inside nearly every app and device, yet much of it is maintained by tiny volunteer teams with no budget. OpenAI is working with Trail of Bits, HackerOne and project maintainers, and more than 30 projects have signed on, including cURL, Go and Python. These are tools billions of devices quietly depend on every day.
The same power cuts both ways. An AI good enough to find and fix flaws is also good enough to find and exploit them. OpenAI is gating the strongest tier, GPT-5.5-Cyber, behind vetted access and monitoring, while steering most users to a lighter version. The bet is that putting scalable defense in trusted hands first tips the balance before attackers get the same capability. That bet is far from settled.
So OpenAI is trying to turn its AI into the world's patch crew. Top benchmark score, more than 20 security partners, and a plan to harden the open-source backbone of the internet. Whether it outruns the attackers wielding the same tools is the real test. Watch how fast Patch the Planet ships actual fixes.