AIUC raises $40M Series A to build frontier AI auditing standards and insurance — modeled on nuclear

AIUC raises $40M to build frontier AI auditing and insurance

Rune Kvist's company AIUC has closed a $40 million Series A to scale what he describes as the critical missing infrastructure for frontier AI deployment: independent auditing standards backed by insurance incentives.

The core argument is that demand for AI audits is no longer the problem. The bottleneck is auditor capacity — finding people who understand both frontier AI systems and what a rigorous audit actually looks like — and, more fundamentally, incentives. Right now, labs choose their own auditors, which Kvist likens to picking your own dog to grade your homework.

“Dario, Sam, Milan, and I are all talking about how do we build trust in the models before we deploy them, how do we build frontier auditing. That's top of mind for us. The short-term bottleneck is on capacity of world-class auditors who both get frontier AI systems and know what an audit looks like. That's where we think insurance plays a really important role — you want to shift from a world where labs pick their own auditors.”

Insurance as the fix for incentive misalignment

Kvist's proposed solution borrows from nuclear energy governance. In nuclear, plant operators are required to carry mutual insurance, meaning they are financially on the hook for each other's failures. That structure turns peers into watchdogs. Kvist wants to replicate that dynamic for AI: shift audit selection from labs to insurers, who have direct financial exposure if they get the risk assessment wrong and therefore have genuine incentive to tell the truth.

He frames this as a commercial approach to AI safety rather than a regulatory one, and explicitly wants to avoid the overcorrection he sees in nuclear's history, where heavy post-incident regulation effectively killed the industry.

The liability gap

Naval Ravikant recently argued that strong liability enforcement — making developers responsible if their agent swarms go rogue or their models get jailbroken — could be sufficient to discipline the market. Kvist broadly agrees with the logic but identifies two gaps.

First, existing legal frameworks weren't designed for AI. Cyber law, for example, typically requires evidence of human intent to hack. An incident involving an autonomous model on a platform like Hugging Face doesn't fit that template cleanly, leaving courts in a gray area.

Second, liability only scales to the size of the liable party's balance sheet. Kvist notes that some estimated AI risk scenarios could exceed even the balance sheets of the largest labs. That's where catastrophe insurance — modeled on tools the industry has developed over decades for low-frequency, high-severity events — becomes relevant.

His proposed bridge is to use the labs' own existing market-cap exposure as the initial insurance layer, capturing the incentive benefits of a formal watchdog without needing to immediately stand up trillions in third-party insurance capacity.

Kvist acknowledges that pricing this risk precisely isn't yet possible — much like pricing the first nuclear plant wasn't — but argues the quantitative tools and frameworks from catastrophic risk modeling are a starting point. The $40M gives AIUC the runway to start building that infrastructure before the question becomes urgent.