OpenAI launches GPT-5.6-Cyber as companies prepare for AI agents hacking

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