Cogent VR-1: AI Model Proves Attack Paths, Halves Cost

Vulnerability management startup Cogent Security has introduced Cogent VR-1, a new reasoning model. This model finds and proves attack paths within live enterprise environments. The company states VR-1 proved twice as many attack paths as other frontier models on IntrusionBench, a new benchmark. It did this at roughly a quarter of the cost. IntrusionBench tests an agent by giving it only a foothold and a target. The agent must then navigate cloud infrastructure, identity systems, and internal tools. Frontier models are good at finding single bugs. However, real intrusions involve chains of small weaknesses. Cogent trained VR-1 to link these weaknesses together, similar to how an attacker would. It then checks if a proposed fix truly closes the gap or just moves the risk. The comparison set for VR-1 included Kimi K3, Claude Opus 4.8, and GLM-5.2. Cogent did not benchmark VR-1 against Mythos. The company is marketing VR-1 as "Mythos-class," pitching it as a defensive tool against autonomous AI attacks. Vineet Edupuganti, Cogent's co-founder and CEO, states VR-1 allows them to quantify AI capabilities in security. He says they can show specific attack paths and then use the same model to close them. The Cogent AI Harness ships with VR-1. This runtime provides environment context, tools, and policy enforcement to any capable model. This development could change how organizations defend against sophisticated cyber threats.