AIUC raises $40M and launches a standardized audit for AI agents

Two former AI security experts launch AIUC, an audit and certification layer for AI agents, and secure $40 million in Series A funding. A dedicated standard, 5,000 tests, and detailed reports aim to help buyers assess reliability and risks, as industry players advocate for independent evaluations.

Third-Party Assessments Requested and METR's Role Mentioned by Anthropic

Dario Amodei, CEO of Anthropic, recently called for a slowdown in the development of cutting-edge technologies due to a rapid increase in incidents of inappropriate behavior. He also mentioned the idea of imposing integrated third-party evaluators to observe and verify safety, citing METR as a possibility. METR is already conducting tests for leading labs, with a long-standing focus on performance, and was consulted by OpenAI during the investigation of its incident related to Hugging Face. In this landscape where the demand for independent evaluations is growing, AIUC positions itself as an audit provider without direct integration with clients, offering independent assessments of agent safety.

An In-House Standard Inspired by SOC 2 and a Battery of 5,000 Tests

AIUC has built a third-party audit and certification layer for AI agents. The company drew inspiration from the SOC 2 standard to create AIUC-1 and a testing service that validates agents against this benchmark. Specifically, each agent undergoes approximately 5,000 tests covering scenarios such as jailbreaks, hallucinations, and data leaks. The results culminate in a report of about 100 pages, detailing where the agent behaves safely and reliably and where it has weaknesses. According to Rune Kvist, the execution of the tests relies on AI agents and data analysis on AI, with human verification of the final audit. The services are not integrated with the client, as AIUC focuses on independent evaluation.

Criteria Dictated by 250 Risk and Security Managers

To develop its standard, AIUC assembled a consortium of around 250 security and risk management leaders identified as buyers of agents. Rajiv Dattani notes that these stakeholders meet monthly to clarify what they are looking for in a purchase, the questions they want to ask, and the issues to address. Their feedback directly informs the design of the testing suite. Dattani explains that the reports clearly distinguish validated points, areas of trust, and concerns to be aware of before any purchasing decision.

What Buyers Are Asking For and Who Is Already Using AIUC

According to Rune Kvist, banks, hospitals, governments, and military organizations are no longer blocking AI due to a lack of model intelligence, but because they have made specific commitments to their clients that no provider can guarantee today. AIUC targets these needs for security and reliability assurance for businesses and agent developers. The company already counts Cursor, Lovable, Harvey, and ElevenLabs among its clients.

Founders, Trajectories, and $55 Million Already Raised

AIUC was founded by Rune Kvist, a former employee of Anthropic, and Rajiv Dattani, former COO of METR from 2024 to 2025, who still sits on its board. Kvist is also Dattani's brother-in-law. The startup announced a Series A funding round of $40 million led by Ribbit Capital, with First Harmonic. It had previously raised $15 million in seed funding led by Nat Friedman through NFDG, with Emergence, Terrain, and Ben Mann, bringing the total to $55 million. Kvist asserts that, paradoxically, AI is becoming more difficult to adopt and control as it gains intelligence, a belief that fuels AIUC's approach. The founders aim for a solution to prevent uncontrollable AI agents in businesses. In the same industry context, Jacob Coxon's departure from Anthropic and his concerns about serious risks by the end of the decade illustrate the current tensions surrounding the safety of systems.