OpenAI is rolling out a specialised, cyber-permissive version of its artificial intelligence (AI) model, dubbed GPT-5.6-Cyber, granting vetted cybersecurity teams access to advanced tools designed to defend against autonomous cyberattacks carried out by AI agents. The release comes shortly after OpenAI paused the launch of its next-generation Astra model after internal safety evaluations revealed the system had achieved critical automated hacking capabilities.
OpenAI launches two Daybreak tiers to unlock high-capability defense tools
Cybersecurity defenders have long faced high refusal rates when using standard commercial AI models, as safety guardrails built to prevent malicious hacking often block legitimate security research.To address this friction, OpenAI is expanding its “Daybreak” initiative, which is a special access program for defensive security teams, into two distinct operational tiers:Daybreak Blue offers access to the base GPT-5.6 Sol model with reduced system-level guardrails, allowing security teams to analyse code vulnerabilities with fewer automated refusals, and second Daybreak Red that grants access to the specialised GPT-5.6-Cyber model, enabling researchers to validate complex exploits, simulate attack chains, and perform advanced vulnerability testing.During internal testing, GPT-5.6-Cyber successfully responded to 95 percent of requests involving complex cybersecurity tasks, including privilege escalation, authentication bypass, and multi-step exploit chains. By comparison, the standard GPT-5.6 Sol model answered just 1.5 percent of similar queries due to built-in safety filters.OpenAI noted that while GPT-5.6-Cyber is significantly more capable than standard models, it met the “High” capability threshold under the company’s Preparedness Framework, unlike Astra, which breached higher risk categories during stress testing.Under the updated Daybreak framework, major cybersecurity and enterprise service providers, including Accenture, IBM, CrowdStrike, Cisco and Palo Alto Networks, can integrate GPT-5.6-Cyber directly into commercial security products, managed threat detection services, and client consultancies.
Click Here For The Original Source.
