Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents

Amid escalating concerns over AI safety, a new startup founded by an early Anthropic employee and the former COO of METR has secured $55 million in total funding to provide enterprises with a verification layer for AI agents. The company, Artificial Intelligence Underwriting Company (AIUC), just announced a $40 million Series A led by Ribbit Capital. Co-founders Rune Kvist and Rajiv Dattani are positioning their firm as a critical solution for enterprises that are hesitant to deploy advanced AI not because of capability, but due to the inability to guarantee safe and predictable behavior.

AIUC has developed a commercial audit and certification service modeled on the established SOC 2 cybersecurity standard. Its own framework, called AIUC-1, was built with input from a consortium of 250 security and risk leaders. The company, which counts Cursor, Lovable, Harvey, and ElevenLabs among its customers, subjects AI agents to a battery of 5,000 automated tests designed to probe for vulnerabilities like jailbreaks, hallucinations, and data leakage. The process culminates in a comprehensive report detailing an agent's performance, providing a clear basis for enterprise adoption decisions. The new funding follows a $15 million seed round from investors including Nat Friedman and Anthropic co-founder Ben Mann.

AIUC's Enterprise Safety Framework

• Standard: Developed the AIUC-1 certification, an AI-specific framework modeled after the widely adopted SOC 2 cybersecurity compliance standard.
• Testing: Utilizes a suite of approximately 5,000 tests, run by other AI agents, to evaluate agent behavior in various scenarios.
• Scope: Audits focus on critical enterprise risks including jailbreaks, model hallucinations, and potential for data leaks.
• Output: Delivers a detailed, 100-page report that outlines where an agent performs safely and identifies areas of concern for potential buyers.
• Validation: While testing and analysis are AI-driven, human experts verify the final audit to ensure accuracy and reliability.

The emergence of AIUC signals a significant shift in the AI safety landscape, moving from theoretical research at frontier labs to a commercially available service for the broader market. While organizations like METR focus on evaluating core frontier models, AIUC is building a scalable, productized solution for the enterprises that will ultimately deploy agents built on those models. This addresses a major adoption barrier cited by Kvist: enterprises need verifiable guarantees that an AI system will adhere to its operational commitments, a gap that AIUC aims to fill by providing independent, standardized assessments.

AIUC's strategy is to productize AI safety, translating the abstract concerns of frontier model research into a concrete, repeatable compliance framework for enterprise buyers. By mirroring the SOC 2 model, the company is betting that trust and third-party verification, not just raw capability, will become the primary drivers for widespread AI agent adoption in high-stakes environments.