AIUC raises $40 million to audit AI agents, and Cursor, Lovable, Harvey and ElevenLabs are already customers

TechCrunch reports that the Artificial Intelligence Underwriting Company, or AIUC, raised a $40 million Series A led by Ribbit Capital, with First Harmonic joining. The company was founded by Rune Kvist, an early Anthropic employee, and Rajiv Dattani, former chief operating officer of the evaluation lab METR. Total funding to date is $55 million, including a $15 million seed round.

What AIUC actually does

AIUC audits AI agents. It does not sell insurance in the ordinary sense, and it does not build the guardrails itself. It ran about 5,000 tests per audit, TechCrunch reports, checking how an agent behaves under jailbreak attempts, hallucination pressure, and data leak conditions. The output is a report of roughly 100 pages that names where the agent held up and where it did not.

The standard the tests measure against is called AIUC-1, which the founders modelled on SOC 2, the cybersecurity attestation many enterprise buyers already ask for before they will sign. The idea is that a buyer looking at three agents can compare three AIUC-1 reports the same way they compare three SOC 2 reports today.

Who is already using it

TechCrunch names four customers: Cursor, Lovable, Harvey, and ElevenLabs. Two of those, Cursor and Lovable, are coding tools that a lot of engineering teams already run inside their workflow. Harvey sells to law firms and ElevenLabs sells voice, so the audit standard is being asked for across quite different agent categories, not just coding.

Dattani, quoted in the piece, described the shape of the report as "Here's where it passes and where you can trust it. And here's where there's concerns." The framing is that an enterprise buyer sees a specific list of behaviours the agent handles well and a specific list it does not, before the buyer has to write a cheque.

The tests use AI, the sign-off does not

The tests themselves run with AI, and AI helps read the results, but TechCrunch reports that a human verifies the final audit. That matters, because the whole point of a third party audit is that a buyer trusts the sign-off, and a fully automated audit puts the reader back to trusting a model's judgement of another model.

For a team about to ship an agent to production, the practical consequence is that agent security is starting to look like the SOC 2 conversation from ten years ago. A prospective customer asked whether the vendor had one, the vendor either did or scrambled to get one, and the scramble ended up shaping what the vendor built. If AIUC-1 gets traction with buyers, an agent vendor's shape-of-work list picks up a new line item, and that line item has a $40 million company writing the questions.

Source

Source: TechCrunch, 15 September 2026.