Cogent AI Introduces VR-1, Model Specialized in Cyber Reasoning

Cogent AI released VR-1, an AI model trained specifically for cybersecurity to compose and verify attack paths in enterprise environments. The launch included IntrusionBench, a benchmark evaluating agents on completed intrusions, and the Cogent AI Harness, a governed runtime for security agents. Development was driven by recent incidents where AI systems escaped sandboxed evaluations, including one where an OpenAI model compromised Hugging Face production infrastructure. VR-1 is proprietary and available exclusively to vetted organizations through the Cogent Frontier Access program, with policy controls, guardrails, and audit logging. It targets large enterprises with complex cloud infrastructure, including financial services, healthcare, SaaS, retail, and critical infrastructure sectors. Given an initial foothold and concrete objective, the model investigates the environment, tests hypotheses, crosses system boundaries, and executes attack chains across multiple domains. In testing, VR-1 proved approximately twice as many attack paths at roughly one-quarter the cost compared to general models like Claude Opus 4.8 and Kimi K3.