OpenAI has expanded its Daybreak cyber defense service with two tiers—Blue and Red—and a new cyber-focused model, GPT-5.6-Cyber. Every day seems to bring fresh news of an AI agent going “rogue,” whether compromising Hugging Face, hacking a gym website, or creating fake profiles to socially engineer an intrusion. The AI labs that make the models doing the hacking are expanding their cyber protection offerings.

OpenAI announced the expansion of Daybreak, a service that bundles access to models, tools, and workflows for defenders. The two new tiers are Blue and Red, both allowing approved customers access to OpenAI’s limited-access frontier cyber models. Frontier models—the most advanced available—have been controversial; the Trump administration previously sought to collaborate with AI companies on their rollout over safety concerns, and OpenAI previously deployed significant guardrails limiting what customers could do with them.

Blue is described as the “recommended starting point for most defenders,” offering incident response, malware analysis, and patch validation. Red provides a broader and potentially more dangerous toolkit, including “purpose-trained cybersecurity models” for security testing and vulnerability research. Red also includes the new GPT-5.6-Cyber model, built off GPT-5.6 Sol, with enhanced capabilities for specialized cybersecurity tasks. GPT-5.6-Cyber is initially available only for “trusted customer partners” such as Accenture, IBM, Crowdstrike, and Cloudflare.

While threats from AI agents are rapidly increasing, critics argue these services function as marketing opportunities for AI labs. OpenAI is marketing Daybreak accordingly: “The cybersecurity world is rapidly changing—threat actors will increasingly use AI to conduct cyberattacks at unprecedented speed and scale, including in fully autonomous ways,” the company wrote. “As these capabilities spread, defenders have a narrowing window to prepare.” Enterprises remain interested in buying protection from AI labs who know the security risks first-hand.