Agents Now Come With a Safety Certificate, and Cursor Already Has One
AIUC raised a $40 million Series A led by Ribbit Capital with First Harmonic participating, announced September 15 at https://techcrunch.com/2026/09/15/early-anthropic-hire-former-metr-coo-have-found-a-way-to-rein-in-rogue-ai-agents/. That sits on top of a $15 million seed from Nat Friedman's NFDG, Emergence, Terrain and Anthropic co-founder Ben Mann, which puts total funding at $55 million. The full name is the Artificial Intelligence Underwriting Company, and the name tells you the whole thesis.
The founders are Rune Kvist, an early Anthropic employee, and Rajiv Dattani, former COO of METR. They are brothers-in-law, which is a fun detail and also a slightly awkward one given what the company sells. What it sells is an audit. You hand AIUC your agent, it runs roughly 5,000 tests against it for jailbreaks, hallucinations and data leaks, and you get back a hundred-page report and a certificate against a standard they wrote called AIUC-1. Cursor, Lovable, Harvey and ElevenLabs are already customers.
Kvist's framing of why anyone pays for this is the sharpest sentence in the whole announcement. Banks, hospitals, governments and militaries are not refusing to deploy agents because the models are too dumb. They are refusing because they made promises to their own customers about what a system will and will not do, and right now nobody on earth can guarantee that. Capability stopped being the bottleneck a while ago. Liability is the bottleneck, and liability is a document problem before it is a model problem.
Here is why this is bigger than one Series A. A METR alum starting a commercial certification body, on the same day [the three big labs confirmed they have been negotiating an industry standards body](https://clauday.com/article/c87f9d35-ee9d-48dd-b2ac-856d5d3adedc), is the evaluation-independence question turning into a market. We have been tracking this thread all week β [one Tel Aviv firm ran the cyber evals behind all three lab escape disclosures](https://clauday.com/article/47ac9624-7e5f-41f9-8ec8-87d13cfa3eab), and [Sacks argued METR is too entangled with Anthropic to audit it](https://clauday.com/article/47053afd-a47a-4663-8395-fc0b8d3f9ad8). The answer nobody liked was "trust the evaluator." The answer AIUC is proposing is "pay a different evaluator, and make it underwrite the result."
The word underwriting is doing real work there. An auditor who is wrong writes a bad report. An underwriter who is wrong pays out. If AIUC actually gets to the insurance end of its own name, that is the first mechanism in this entire debate where being wrong about agent safety costs somebody money instead of reputation. Watch whether AIUC-1 gets cited by a procurement department that is not a startup.
← Back to all articles
The founders are Rune Kvist, an early Anthropic employee, and Rajiv Dattani, former COO of METR. They are brothers-in-law, which is a fun detail and also a slightly awkward one given what the company sells. What it sells is an audit. You hand AIUC your agent, it runs roughly 5,000 tests against it for jailbreaks, hallucinations and data leaks, and you get back a hundred-page report and a certificate against a standard they wrote called AIUC-1. Cursor, Lovable, Harvey and ElevenLabs are already customers.
Kvist's framing of why anyone pays for this is the sharpest sentence in the whole announcement. Banks, hospitals, governments and militaries are not refusing to deploy agents because the models are too dumb. They are refusing because they made promises to their own customers about what a system will and will not do, and right now nobody on earth can guarantee that. Capability stopped being the bottleneck a while ago. Liability is the bottleneck, and liability is a document problem before it is a model problem.
Here is why this is bigger than one Series A. A METR alum starting a commercial certification body, on the same day [the three big labs confirmed they have been negotiating an industry standards body](https://clauday.com/article/c87f9d35-ee9d-48dd-b2ac-856d5d3adedc), is the evaluation-independence question turning into a market. We have been tracking this thread all week β [one Tel Aviv firm ran the cyber evals behind all three lab escape disclosures](https://clauday.com/article/47ac9624-7e5f-41f9-8ec8-87d13cfa3eab), and [Sacks argued METR is too entangled with Anthropic to audit it](https://clauday.com/article/47053afd-a47a-4663-8395-fc0b8d3f9ad8). The answer nobody liked was "trust the evaluator." The answer AIUC is proposing is "pay a different evaluator, and make it underwrite the result."
The word underwriting is doing real work there. An auditor who is wrong writes a bad report. An underwriter who is wrong pays out. If AIUC actually gets to the insurance end of its own name, that is the first mechanism in this entire debate where being wrong about agent safety costs somebody money instead of reputation. Watch whether AIUC-1 gets cited by a procurement department that is not a startup.
Comments