Cogent AI Team Releases VR-1: A Frontier Cyber Reasoning Model That Composes and Verifies Enterprise Attack Paths

Cogent AI has launched VR-1, a cyber reasoning model tuned for enterprise intrusion workflows rather than general coding.

The release also includes IntrusionBench for execution-based attack-path scoring and the Cogent AI Harness for controlled security-agent runtime.

VR-1 is aimed at large enterprises and government users, with access limited to vetted organizations through a frontier access program.

Cogent says VR-1 outperformed several competing models on black-box attack-path completion, while stressing it was evaluated on attack-chain composition only.

The launch highlights both the value of better defensive testing and the risk of more capable offensive automation in high-stakes security settings.