OpenAI launches GPT-5.6-Cyber model to enhance vulnerability discovery and security research capabilities.

OpenAI announced an expansion of its Daybreak cybersecurity defense initiative, alongside the launch of the GPT-5.6-Cyber model—a purpose-built AI for cybersecurity designed to help authorized security researchers and defense teams boost their capabilities in vulnerability discovery, threat analysis, and security testing. As adversaries increasingly leverage AI to launch faster, more large-scale cyberattacks, defenders need access to advanced AI capabilities to stay ahead of threats. Daybreak now offers two tiers of access: Daybreak Blue, for most defense teams, provides general-purpose models including GPT-5.6 Sol, for use cases such as vulnerability discovery, security code review, malware analysis, incident response, and patch validation. Daybreak Red, meanwhile, is geared toward advanced security research, offering GPT-5.6-Cyber for authorized vulnerability research, verification, and security testing. GPT-5.6-Cyber is trained on GPT-5.6 Sol and optimized specifically for cybersecurity tasks, including zero-day vulnerability discovery and exploit chain analysis. The model has already been deployed in real-world vulnerability research, identifying critical software flaws including those in Chrome’s V8 JavaScript engine. Additionally, OpenAI noted that GPT-5.6-Cyber has aided in uncovering critical vulnerabilities across multiple domains: privilege escalation bugs in mobile operating systems, remote code execution (RCE) flaws in databases, and hundreds of privilege escalation vulnerabilities in operating system kernels. OpenAI also stressed that Daybreak Red will only be available to pre-approved individuals and organizations, with access controlled via measures such as identity verification, account security protocols, monitoring, usage restrictions, and legal disclaimers. The company said it will continue to strengthen security monitoring, access management, and model safety testing to mitigate the risk of advanced cybersecurity models being misused.