OpenAI has introduced GPT-6 Cyber, a specialised model for authorised security research, and says it is better both at developing working exploits and at staying inside the boundaries researchers set. The company is starting with a selected group of approved Daybreak Red customers, rather than releasing it to everyone.

Rows of server racks in a data centre aisle, an illustrative cyber infrastructure file photo, not OpenAI's offices. Photo: The National Archives UK / Wikimedia Commons, CC BY 3.0, modified.