Smarter AI is harder to control

Rune Kvist, an early Anthropic employee, and Rajiv Dattani, former COO of safety org METR, built a company on one counterintuitive premise about AI capability.

“It becomes harder to adopt and harder to control as AI gets smarter, not easier”

A $55M bet on safety

Their startup AIUC announced a $40 million Series A led by Ribbit Capital, on top of a $15 million seed from Nat Friedman's fund and Anthropic co-founder Ben Mann.

Cursor and Lovable signed on

AIUC already names Cursor, Lovable, Harvey, and ElevenLabs as customers, companies that build the agents everyone else is trying to deploy.

The real blocker is promises

Banks, hospitals, governments and militaries are not held back by model intelligence. They are held back by commitments they made to their own customers.

“They decline because they've made commitments to their own customers about what a system will and won't do, and nobody can currently guarantee that.”

SOC 2, but for agents

AIUC copied the playbook of the cybersecurity certification everyone already trusts. It built a standard called AIUC-1 plus a testing service that validates agents against it.

250 buyers wrote the test

Rather than inventing criteria, AIUC convened about 250 security and risk leaders and asked what they actually check before buying an agent. Their answers shape the tests.

“When you're buying agents from someone, what would you look for?”

5,000 tests, one 100-page report

Agents get run through jailbreaks, hallucination traps, and data leak scenarios. AI runs the tests and analyzes results, but humans verify the final audit.

Third-party eyes are coming anyway

Dario Amodei has floated embedding independent evaluators inside frontier labs, naming METR as an option. AIUC sells the same assurance to the enterprises buying the agents.

“Here's where it passes and where you can trust it. And here's where there's concerns.”