This startup wants to give AI agents a safety report card before they reach your bank or hospital

Key points

• AIUC, the Artificial Intelligence Underwriting Company, announced a $40 million Series A led by Ribbit Capital on Tuesday.
• The startup has raised $55 million in total, including a $15 million seed round backed by Anthropic co-founder Ben Mann and others.
• AIUC runs each AI agent, software that can carry out multi-step tasks on its own, through roughly 5,000 tests covering jailbreaks, hallucinations and data leaks.
• Customers already signed on include Cursor, Lovable, Harvey and ElevenLabs.
• Co-founder Rajiv Dattani served as COO of METR, an AI safety research organisation, from 2024 to 2025.

Banks and hospitals aren't turning down AI because it's too dumb. They're turning it down because nobody can tell them exactly what it will do, or refuse to do, once it's running inside their systems.

That's the problem Rune Kvist and Rajiv Dattani are trying to solve. The pair, who are also brothers-in-law, founded the Artificial Intelligence Underwriting Company, known as AIUC, to bring independent safety audits to AI agents before those agents get deployed at scale.

Kvist was an early employee at Anthropic, the AI safety company behind the Claude family of chatbots. Dattani spent 2024 to 2025 as COO of METR, a non-profit that tests frontier AI models, meaning the most powerful and capable AI systems currently being built, for dangerous behaviours. He remains on METR's board.

How does the audit actually work?

AIUC developed its own standard, called AIUC-1, using input from about 250 security and risk leaders at the kinds of organisations that buy and deploy AI agents. Every month, the company asks those leaders what they would need to see before trusting an agent with their systems.

That feedback shapes a testing suite of roughly 5,000 individual checks. Agents face scenarios designed to trigger jailbreaks (attempts to make AI ignore its safety rules), hallucinations (when AI confidently states something false) and data leaks. Results land in a report of around 100 pages that tells a buyer plainly where an agent behaves safely and where it doesn't. AI agents run the tests and analyse the data; a human auditor verifies the final report.

The model borrows from SOC 2, a widely recognised cybersecurity compliance standard that companies already use to show they handle customer data responsibly. Get an independent audit, get a certificate, let buyers make informed decisions.

What does this mean if you work at a company buying AI tools?

For now, the certification is voluntary. No law requires it. But for anyone in procurement or legal at a large business, an AIUC audit report gives you something concrete to put in front of your risk team: here is what this agent will do, and here is what it won't.

Dattani told TechCrunch: "Here's where it passes and where you can trust it. And here's where there's concerns. You should be aware of those references before you make the decision to buy."

The $40 million Series A, first reported by TechCrunch AI, was led by Ribbit Capital with participation from First Harmonic. The earlier $15 million seed came from Nat Friedman through his fund NFDG, alongside Emergence and Terrain, as well as Anthropic co-founder Ben Mann.

Anthropic's CEO Dario Amodei recently called for frontier labs to use embedded third-party evaluators to verify AI safety, naming METR as one option. We covered that push on 12 September. AIUC isn't proposing to embed itself inside customer organisations, but the direction is close: independent eyes on AI behaviour before something goes wrong. Given what we've reported about agents breaking containment and attacking outside systems, a voluntary audit scheme may be the only structured check enterprises have right now.