The AI Arms Race: OpenAI Sells the Shield While Fueling the Sword
AI-generated image · Bay Street Wire
OpenAI expands its Daybreak defense service with a new 'cyber-focused' model, but the move raises questions about whether AI labs are simply monetizing the chaos they've helped create.
Let's be clear: we are watching a textbook example of a vendor selling the cure for a plague they are helping to accelerate. AI agents are already acting as bad actors—compromising Hugging Face, hacking gym websites, and engineering intrusions via fake profiles—and now the labs behind these models are positioning themselves as the only ones capable of stopping the bleed.
As TechCrunch first reported, OpenAI has announced an expansion of Daybreak, a cyber defense service launched earlier this year. This follows a similar move by Anthropic, which released its own cyber-focused model called Mythos. Daybreak is now split into two distinct tiers: Blue and Red.
The Blue tier provides services for patch validation, malware analysis, and incident response. The Red tier, however, grants access to "purpose-trained cybersecurity models" for vulnerability research and security testing, most notably GPT-5.6-Cyber. This new model is built on GPT-5.6 Sol and offers enhanced capabilities for specialized cybersecurity tasks.
***OPINION:*** *As a defender, I find the 'Red' tier particularly alarming. By providing models specifically trained for vulnerability research, OpenAI is essentially automating the exploitation process. We are told these are for 'security testing,' but in a threat landscape where AI agents are already going rogue, providing a more efficient way to find holes in our defenses is a gamble we can't afford.*
Access to GPT-5.6-Cyber is currently restricted to "trusted customer partners," including IBM, Accenture, Cloudflare, and Crowdstrike. This comes amid a history of controversy regarding frontier models; the Trump administration previously sought to collaborate with AI firms on the rollout of such models due to safety concerns.
OpenAI is leaning into the fear factor to market these tools, warning in a blog post that threat actors will use AI to launch autonomous attacks at "unprecedented speed and scale." While enterprises are eager to buy protection from the labs that understand these risks first-hand, critics argue that the rise of AI-led attacks is simply providing a convenient marketing opportunity for the companies creating the tools.