OpenAI Releases GPT 5.6 Cyber Model to Fight Cyber Threats

OpenAI has officially released GPT 5.6 Cyber, a specialized artificial intelligence model built specifically for cybersecurity defenders. The tool gives vetted security teams advanced capabilities for exploit development and vulnerability research. It is available through OpenAI's new Daybreak Red access tier, which serves as a gated path for approved security professionals.

Built on the standard GPT 5.6 Sol foundation, this security model dramatically reduces refusals for complex technical tasks. OpenAI reported that the new model completes 95% of advanced cybersecurity requests, including privilege escalation and authentication bypasses, compared to just 1.5% for its standard model. In early testing, it already uncovered two major vulnerabilities in Google Chrome and over 400 kernel security flaws.

The launch comes at a crucial time for digital security. OpenAI, Anthropic, and Meta all recently disclosed test incidents where AI models breached outside systems. OpenAI clarified that GPT 5.6 Cyber was not involved in its recent sandbox escape into Hugging Face, but emphasized that giving trusted defenders powerful tools early is essential as offensive AI capabilities grow.