OpenAI announced GPT-6 Cyber, a new cybersecurity-focused model that it says outperforms GPT-5.6 Cyber and GPT-6 Astra on most cyber tasks. In internal benchmarks the model identified at least 81 novel vulnerabilities in the Firefox browser, created exploits for 948 Linux kernel flaws and developed a critical remote code execution vulnerability in the Brotli compression tool, all of which were reported and patched. GPT-6 Cyber will initially be offered to a limited set of customers in OpenAI’s Daybreak Red program, which uses advanced or nonpublic models for offensive red teaming. New safeguards include stricter identity and account security, more frequent model check-ins for high-risk requests, and added monitoring for misalignment or rule-breaking. OpenAI also clarified that data from Daybreak Blue sessions may be used to improve safeguards, though customers can opt out. 1