OpenAI launches GPT-6 Cyber for cybersecurity, under 1% task deviations
Technology
OpenAI just released GPT-6 Cyber, its latest AI built to make the internet safer.
This new model is much better at spotting security threats and cuts down on risky behavior: during tests, it strayed from its tasks less than 1% of the time (way better than the previous version's 5%).
GPT-6 Cyber tops GPT-6 Astra tests
GPT-6 Cyber outperformed earlier models and even beat OpenAI's flagship Critical-level model, GPT-6 Astra, in security tests, plus it uncovered 81 new Firefox bugs.
It also comes with stronger protections to keep bad actors out (think extra safety for things like power plants).
Right now, only select users in OpenAI's Daybreak Red program can try it, but more people will get access soon.
The focus is strictly on using this tech for legitimate cybersecurity defense.