GPT-5.6-Cyber unlocks advanced security research

OpenAI is turning controlled access into a core part of cyber-model deployment, pairing substantially greater capability with strict verification and monitoring.

• –Completes 95% of advanced cyber requests in OpenAI’s internal evaluation, versus 1.5% for standard GPT-5.6 Sol
• –Built for exploit-chain development, authentication bypass research, privilege escalation, and zero-day discovery
• –Daybreak Red restricts access to approved organizations conducting authorized security work
• –The model is specialized rather than universally superior, underperforming GPT-5.6 Sol on some vulnerability-reporting and exploit-efficiency evaluations
• –The real test is whether faster discovery translates into validated patches instead of simply producing more security findings