OpenAI Opens GPT-6 Cyber to Select Customers in Limited Release

OpenAI has begun a restricted rollout of GPT-6 Cyber, a cybersecurity research model that the company says outperforms GPT-6 Astra on most security tasks, with access set to widen gradually.

The model is currently available to some Daybreak Red customers. Security teams at Mozilla, Palo Alto Networks and Cisco are already using it.

OpenAI describes GPT-6 Cyber as more capable and better aligned. The company credits it with finding at least 81 previously unknown, high-risk Firefox vulnerabilities and developing usable exploits for 948 Linux kernel vulnerabilities.

OpenAI also said the model carried out earlier Linux kernel vulnerability research. The findings span browser and operating system security work, from identifying software weaknesses to creating code that exploits them. The staged release gives a limited group of customers access first, with OpenAI planning to expand availability over time.