AIUC raises $40 million to audit rogue AI agents

AIUC is trying to do for AI agents what SOC 2 did for software vendors: give buyers a recognizable seal that says an independent party checked how a system behaves. The standard was shaped by a consortium of about 250 security and risk leaders — described in the article as the buyers of agents — who were asked what they would want to see addressed when purchasing agents.

The startup's leadership gives it direct ties to the AI safety world. Kvist was an early Anthropic employee, and Dattani was COO of METR, the AI safety research organization that has done similar testing for frontier labs, mostly focused on whether agents can reliably complete tasks. Anthropic CEO Dario Amodei has recently called for frontier labs to use embedded third-party evaluators to observe and verify safety and named METR as one possibility. AIUC is not proposing to embed itself at customer sites, but the underlying idea is similar.

The test is whether AIUC-1 becomes something enterprises actually ask for when buying agents, rather than another optional report. That likely hinges on whether the consortium members who helped write the standard treat it as a purchasing requirement, and on whether AIUC's roughly 5,000-test, roughly 100-page reports give buyers the confidence they currently lack.