Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents
TechCrunch Julie Bort ● Covered by 71 sources
Former Anthropic and METR leaders just raised $40M for AIUC, a startup that audits AI agents. It wants to bring SOC 2-style checks to rogue AI before enterprises deploy it.
Based on reporting by TechCrunch, Julie Bort — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A former Anthropic hire and a former METR COO are betting that the way to stop bad AI behavior is not to slow the models down, but to inspect them more like software that handles money, customer data, or government work. Rune Kvist and Rajiv Dattani have launched AIUC, short for Artificial Intelligence Underwriting Company, with a pitch aimed at enterprises and at companies building AI models and agents.
The startup announced a $40 million Series A on Tuesday, led by Ribbit Capital with First Harmonic participating. That follows a $15 million seed round from Nat Friedman’s NFDG, Emergence, Terrain, and Anthropic co-founder Ben Mann, among others, bringing total funding to $55 million. AIUC already counts Cursor, Lovable, Harvey, and ElevenLabs as customers.
What makes the company interesting is the frame it borrows from cybersecurity. AIUC has built a third-party audit and certification layer around a standard it calls AIUC-1, using SOC 2 as the rough model. The company says it pulled together about 250 security and risk leaders to shape the standard, then turned that feedback into a testing service that runs agents through about 5,000 scenarios covering jailbreaks, hallucinations, and data leaks.
The output is a report that runs to roughly 100 pages and spells out where an agent behaves safely and where it doesn’t. AIUC says it uses AI agents to run the tests and AI to analyze the data, but humans verify the final audit. That last step matters, because the pitch here is not “trust the machine.” It is “trust the machine, but only after someone has checked the machine’s homework.”
Dattani’s background makes the comparison to METR hard to miss. He was COO there from 2024 to 2025 and still sits on the board, and METR has done similar testing for frontier labs, though it has mostly focused on whether agents can complete tasks reliably. Anthropic CEO Dario Amodei has also recently argued for slower frontier development and suggested third-party evaluators could help watch for safety problems. AIUC is offering a more enterprise-friendly version of that idea: independent proof before deployment, not after the incident report.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI safety pitch that sounds like it was written by people who’ve actually met procurement teams. The industry keeps pretending trust is a vibe; enterprise buyers keep asking for receipts. SOC 2 for agents is a much better business than another grand sermon about alignment from the usual priesthood.
Read more about this at: TechCrunch