Cogent

Building @Cogent_security | prev @AbnormalSec @join_royal @DukeU

Cogent VR-1 is a cybersecurity reasoning model aimed at enterprise security teams who are drowning in individual vulnerability findings but lack the time to figure out which ones actually matter. Given a foothold inside a customer environment, Cogent Security Inc. today introduced Cogent VR-1, a frontier reasoning model trained to find and prove attack paths inside live enterprise environments . The model chains weaknesses across code, cloud, identity, endpoints, and applications to show how an attacker could actually move through a specific environment, then checks whether a proposed remediation truly closes the path or just shifts it elsewhere. The company is pairing VR-1 with the Cogent AI Harness, a governed runtime that supplies environment context, scoped tools, policy controls, and verification , which is how the model grounds its reasoning in a given organization's crown jewels, identities, and runtime controls rather than reasoning about security in the abstract. Access is being rolled out through the Cogent Frontier Access Program, a limited research initiative for qualified enterprises and defenders, positioning this launch as a targeted rollout rather than general availability. The timing tracks with a broader industry shift where autonomous AI attacks have started hitting real targets and time to exploit has compressed sharply. Cogent was founded by Vineet Edupuganti, CEO; Geng Sng, CTO; and Thanos Baskous , with Edupuganti and Sng previously worked at Abnormal Security, where Edupuganti led product strategy and Sng built a machine learning fraud detection system used by a large share of the Fortune 500. VR-1 is a natural extension of the company's earlier work on agentic vulnerability management, moving from prioritizing individual findings to reasoning about the full multi-step chains that security teams struggle to reconstruct by hand.

Tags

Comments (14)