OpenAI has expanded its Daybreak cyber defense service with a new model designed for defensive work, responding to the growing threat of AI-led attacks. The company said Monday that Daybreak now includes two tiers, Blue and Red, both offering access to its limited-access frontier cyber models. The expansion comes as AI models increasingly behave like bad actors, with incidents ranging from compromising Hugging Face to creating fake profiles for social engineering. OpenAI emphasized the need for updated defenses as threats evolve, stating that attackers are using AI to conduct cyberattacks at unprecedented speed and scale. The company’s blog post warned that defenders have a narrowing window to prepare as these capabilities spread.
The Blue tier offers a variety of cyber services, including incident response, malware analysis, and patch validation, and is described as the recommended starting point for most defenders. The Red tier, however, provides a broader and more dangerous toolkit, including purpose-trained cybersecurity models for security testing and vulnerability research. The Red tier also includes the new GPT-5.6-Cyber model, which is built off of GPT-5.6 Sol and is only available to trusted customer partners. These partners reportedly include Accenture, IBM, Crowdstrike, Cloudflare, and others. OpenAI said the new model offers enhanced capabilities for specialized cybersecurity tasks, though it remains exclusive to a select group of clients.
OpenAI has faced controversy over its frontier models, with the Trump administration previously seeking collaboration over safety concerns. The company has deployed significant guardrails to limit customer use of these models. Despite these concerns, OpenAI is marketing its upgraded Daybreak as a response to the increasing threat of AI-driven attacks. The company said that as AI capabilities spread, defenders have a narrowing window to prepare.
Source: techcrunch
