Jacob Coxon’s dramatic exit from Anthropic highlighted a growing panic: smart AI might become uncontrollable. While existential dread grabs headlines, founders Rune Kvist and Rajiv Dattani are tackling a more immediate corporate headache. They argue that as models get smarter, they actually become harder to adopt because enterprises can’t guarantee behavior. Their solution? Artificial Intelligence Underwriting Company (AIUC), a startup that treats AI safety like financial underwriting.
The market seems to agree with their approach. AIUC just closed a $40 million Series A led by Ribbit Capital, bringing its total funding to $55 million. The round included heavy hitters like Nat Friedman and Anthropic co-founder Ben Mann. Early customers already include Cursor, Lovable, Harvey, and ElevenLabs. These companies aren’t just buying tech; they’re buying trust. Banks and hospitals don’t reject AI because it’s dumb. They reject it because they can’t promise clients it won’t hallucinate or leak data.
To solve this, AIUC built AIUC-1, a certification standard modeled after the cybersecurity staple SOC 2. They didn’t guess what buyers want. Instead, they asked 250 security and risk leaders what they’d check before signing a contract. That feedback shaped a battery of 5,000 automated tests covering jailbreaks, hallucinations, and data leaks. The system uses AI agents to run the gauntlet, but humans verify the final audit. The result is a roughly 100-page report that spells out exactly where an agent is safe and where it’s risky.
This model mirrors the work of METR, where Dattani previously served as COO. While METR focuses on frontier labs and performance, AIUC targets enterprise deployment. It aligns with Anthropic CEO Dario Amodei’s recent call for third-party evaluators to pace development. By providing an independent scorecard, AIUC lets buyers see the cracks in the foundation before they build on it. It’s not a cure for rogue AI, but it’s a necessary seatbelt for the ride.
