Cogent Launches AI-Driven Defense Against Autonomous Attack Swarms

The launch arrives as automated threats escalate in sophistication and volume. In July, a swarm of 700 AI agents targeting Hugging Face performed over 17,600 actions, eventually chaining disparate vulnerabilities and service-account tokens to reach 136 production keys. While human attackers prioritize high-yield targets, agent swarms pursue deeper, less obvious paths that human teams often overlook. Cogent research indicates that AI-viable attack paths are twice as deep as those accessible to humans, with the average enterprise seeing a 386% increase in these risks over the past year.

Cogent’s platform aggregates data from existing security tools—including EDR, firewalls, and cloud providers—to construct a live graph of assets and permissions. By applying its VR-1 reasoning model, the system identifies chokepoints where a single, surgical fix can dismantle complex attack chains. According to CEO Vineet Edupuganti, the shift is necessary because criminal groups will soon possess the same autonomous capabilities as research labs, rendering traditional, manual defensive approaches insufficient against non-tiring, relentless agent swarms.