Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents

Rune Kvist, an early Anthropic employee, and Rajiv Dattani, the former COO of the AI safety research organization METR, have built a company that offers independent audits and certification for AI agents used in enterprises. They argue that as AI capabilities advance, control and predictable behavior become harder to guarantee, which is prompting many organizations to hesitate before deploying agents in sensitive contexts. AIUC has created a standard called AIUC-1, modeled on widely adopted cybersecurity frameworks like SOC 2, and assembled a consortium of roughly 250 security and risk leaders to inform what the standard should cover. The startup runs each agent through about 5,000 tests that probe for jailbreaks, hallucinations, and data leaks, then produces a roughly 100-page report that highlights where an agent is reliable and where risks remain. The company says it leverages AI agents to execute the test suites and to analyze results, with humans performing the final verification of audits. AIUC lists Cursor, Lovable, Harvey, and ElevenLabs among its customers, and its approach attracted a Series A led by Ribbit Capital and participation from First Harmonic. AIUC’s offering echoes work being done by METR — where Dattani was COO from 2024 to 2025 and remains on the board — which has performed independent evaluations for frontier labs, historically focused on performance. The wider industry debate about pacing and third-party evaluation has intensified recently; Anthropic CEO Dario Amodei has urged slowing frontier development and suggested embedded third-party evaluators as one way to verify safety, naming METR as an example. AIUC does not propose embedding evaluators on site but aims to give enterprises an independent assessment they can use when deciding whether to buy or deploy an agent.