OpenAI launches two-tier security access programme

OpenAI has launched GPT-5.6-Cyber, a large language model purpose-trained for cybersecurity tasks, and introduced two new tiers to its Daybreak access programme: Daybreak Blue for defensive work and Daybreak Red for advanced defensive and offensive tasks.

Daybreak Blue members get access to frontier general-purpose models like GPT-5.6 Sol with system-level safeguards removed, enabling tasks such as vulnerability discovery, secure code review, malware analysis, incident response and patch validation. Daybreak Red members additionally get access to purpose-trained cyber models like GPT-5.6-Cyber for vulnerability research, exploit validation and security testing.

OpenAI said GPT-5.6-Cyber completed 95% of a set of sensitive requests involving exploit-chain development, authentication bypass and privilege escalation, compared with 1.5% for general-access GPT-5.6 Sol, 2.0% for Daybreak Blue access, and 57.3% for the previous GPT-5.5-Cyber model. The company also used GPT-5.6-Cyber to find CVE-2026-15903, a high-severity vulnerability in Chrome’s V8 engine, which was reported to Google and fixed.

Alex Goller, principal solution architect for EMEA at Illumio, called the two-tier system ‘a good first step’ but noted that guardrails are not the control plane for defence. He said controls should follow zero trust principles, with visibility and segmentation to contain the blast radius when agents operate inside an environment.