As AI-led attacks multiply, OpenAI launches a new cyber model
TechCrunch Lucas Ropek ● Covered by 7 sources
OpenAI just expanded Daybreak with a new cyber model for defenders. It’s split into Blue and Red, and the stronger one is only for trusted partners.
Based on reporting by TechCrunch, Lucas Ropek — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI models are getting pulled into more and more real-world abuse, from compromised accounts to fake profiles built for social engineering. OpenAI is responding the way these labs usually do: by selling the cure alongside the problem. This week it expanded Daybreak, its cyber defense service, and folded in a new defensive model built for security work.
Daybreak now comes in two tiers, Blue and Red. Blue is the softer entry point, with incident response, malware analysis and patch validation. OpenAI says it is the “recommended starting point for most defenders,” which is a polite way of saying most customers probably won’t need the heavier gear.
Red is the sharper tool. It gives approved users “purpose-trained cybersecurity models” meant for security testing and vulnerability research, and it is also where OpenAI has placed GPT‑5.6‑Cyber. That model is built on GPT‑5.6 Sol and is aimed at specialized cybersecurity tasks.
Access is still tightly controlled. OpenAI says both tiers are for approved customers, and GPT‑5.6‑Cyber is only available to “trusted customer partners,” including reportedly Accenture, IBM, Crowdstrike, Cloudflare and others. That keeps the most powerful stuff behind a gate, even as the company argues that attackers are moving toward “unprecedented speed and scale,” including fully autonomous attacks.
The bigger story here is not just the model. It’s that the security market is becoming another place where the AI labs can sell their own expertise back to enterprise buyers. Critics have noticed that too, which is why every new defensive launch also doubles as a very convenient sales pitch.
My take — AI-written commentary, not fact-checked reporting
This is the neat little loop AI keeps building for itself: make the weapon, then sell the shield. OpenAI is not wrong that defenders need help, but the industry should be a lot less impressed when the same shop that ships the risk shows up with a subscription plan. Trusting the labs with security makes sense; trusting them to be humble about it, less so.
Read more about this at: TechCrunch