OpenAI adds GPT-5.6 Cyber to Daybreak Red

OpenAI expanded its Daybreak cybersecurity program with a new Red tier built around GPT-5.6 Cyber, a purpose-trained model derived from GPT-5.6 Sol for security testing and vulnerability research. Access is restricted to trusted partners, reportedly including Accenture, IBM, CrowdStrike, and Cloudflare, not a general API release. Daybreak’s existing Blue tier, aimed at most enterprises, continues to cover incident response and malware analysis.

OpenAI framed the expansion as a response to a run of AI-led attacks: incidents where AI agents compromised Hugging Face, hacked a gym website, and generated fake profiles for social engineering. The company says threat actors will increasingly use AI to run attacks faster and at greater scale, and that Daybreak Red is meant to keep pace on the defensive side.

What it means for you

Most organizations won’t get direct access to GPT-5.6 Cyber; this is a partner-gated model, not a product you sign up for. What you can act on is Daybreak Blue if incident response or malware triage is part of your stack, and the broader signal that OpenAI is now willing to build and gate an offense-capable security model rather than avoid the category. That fits the pattern covered in OpenAI expands Daybreak security tools back in June, and it’s a reminder that if your team runs agents on your security backlog, the attackers you’re defending against are increasingly using the same class of tooling.