OpenAI offers GPT-6 Cyber to approved customers only, rated Critical
The model is built to find and test software flaws with permission. It is not in ChatGPT but in the API and Codex for Daybreak Red customers.
OpenAI has an AI model that finds security mistakes in software. Only approved experts can use it for now. The main thing for you is to keep your phone and computer updated.
Hype check
Louder than it deserves. The real change is smaller.
Our editorial opinion, based on the sources below. How we score
Why it matters
A strong cybersecurity model can help specialists close flaws, which is why OpenAI limits who can use it.
Skip: nothing changes for you; just keep your system and apps updated.
The details
- OpenAI announced it on November 5, and its Preparedness Framework rates it Critical for cybersecurity, the highest tier.
- It is not in ChatGPT. It starts in the and Codex for approved customers of the Daybreak Red program for authorized security testing.
- It completed 48.5% of ExploitGym tasks, against 32.1% for the earlier GPT-5.6 Cyber.
- OpenAI reported 81 previously unknown high-severity flaws in Firefox, and Mozilla credits it for 41 of the 167 fixed in Firefox 159.
- In the Linux kernel it reproduced 1,724 vulnerabilities and built working exploits for 948, and it reported 304 critical findings to maintainers with draft patches.
- Safeguards include IP allowlists and blocks on help with attacks on power plants, water plants and industrial control systems.
Receipts
One trusted outlet reports it. Not confirmed yet. Open the sources and check us.
Written with the help of AI tools and checked against the sources above. How we work
See a mistake? Tell us and we’ll fix it.
