OpenAI released GPT-6 Cyber through API and Codex to a small group of approved Daybreak Red security customers.

OpenAI announced GPT-6 Cyber as a security model trained to identify and validate exploitable vulnerabilities while staying within researcher-defined scope. Initial access is limited to a small group of approved Daybreak Red customers through the API and Codex, with broader Daybreak Red access described as a planned rollout.

A Financial_Express report citing The Verge gives an ExploitGym score of 48.5%, compared with 42.4% for GPT-6 Astra and 32.1% for GPT-5.6 Cyber. The same report says GPT-6 Cyber helped identify 81 previously unknown Firefox vulnerabilities.

Reported out-of-scope execution in testing fell to less than 1%, compared with approximately 5% for the preceding generation. The report also describes strengthened misuse safeguards for high-risk attacks against critical infrastructure, including power plants and water facilities, without providing a quantified safeguard success rate.

GPT-6 Cyber received the highest reported cybersecurity risk classification, “critical,” under OpenAI’s framework. OpenAI also announced availability of GPT-5.6 Cyber to all Daybreak Red customers; that broader access should not be confused with the narrower initial GPT-6 Cyber release.