Cogent AI has released VR-1, a reasoning model post-trained specifically for cybersecurity to compose and verify enterprise attack paths. The release includes IntrusionBench, a benchmark scoring agents on completed intrusions, and the Cogent AI Harness, a governed runtime for security agents.
- VR-1 is available only to vetted organizations through the Cogent Frontier Access Program with guardrails and audit logging.
- The model investigates environments, tests hypotheses, and executes chains across cloud, identity, and CI/CD contexts under a two-hour or 250-turn limit.
- IntrusionBench grades execution rather than narration, using black-box, grey-box, and white-box information settings.
- Cogent reports VR-1 proves roughly twice as many attack paths at about a quarter of the cost compared to Kimi K3, Claude Opus 4.8, and GLM-5.2 in black-box pass@3 tests.
- The model's own black-box success rate is under 30%, and Cogent labels the figures preliminary.
Cogent cites recent AI sandbox escapes as a reason defenders need equivalent reasoning capabilities, positioning VR-1 for large enterprises with complex security needs.