OpenAI launches GPT-6 Cyber for security research with tighter alignment
Briefing by News4U · First published
Today OpenAI said on X that it is introducing GPT-6 Cyber, a major advance in security research that pairs stronger capabilities with better model alignment; that the model powered Linux kernel research announced earlier and is already used by security teams at Mozilla, Palo Alto Networks, Calif, Cisco and OpenAI; and that GPT-6 Cyber is available to select Daybreak Red customers with access expanding over time. The Verge and The Register reported OpenAI classifies the model as Critical under its Preparedness Framework, that it scored 48.5% on ExploitGym versus 32.1% for GPT-5.6 Cyber, and that OpenAI said it avoided exploiting out-of-scope services over 99% of the time in a separate alignment evaluation. [1] [2] [3]
Sources
- @openai (X): Introducing GPT‑6 Cyber: a major advance in security research that pairs stronger capabilities with better model alignment.
@openai (X) · · Excerpt
- OpenAI says its new cybersecurity model is better at not hacking other companies
The Verge · · Full text
- The Register: it avoided exploiting out-of-scope services over 99 percent of the time
The Register · · Excerpt