A Startup Is Building a Safety Certification for AI Agents. It Just Raised $40 Million to Prove the Idea Works.
AIUC, founded in 2024, says its testing standard already covers some of the most widely used AI tools in corporate settings. Now it has the capital to push into larger, more powerful models.
Key points
- AIUC raised $40 million in a Series A round led by Ribbit Capital in 2025, bringing its total funding to $55 million.
- The San Francisco startup, founded in 2024, built AIUC-1, a safety certification standard that runs AI agents through roughly 5,000 simulated attack and failure scenarios.
- Cursor, ElevenLabs, Harvey, KPMG, Lovable, UiPath and Fin have all used AIUC-1 to certify their AI agents.
- AIUC will use the new money to extend its auditing and insurance work to frontier models, meaning the largest, most capable AI systems now entering enterprise use.
Certifying that an AI agent is safe enough to work inside a company sounds like a niche problem. It's not.
AIUC, short for Artificial Intelligence Underwriting Company, announced a $40 million Series A round this week, led by Ribbit Capital with participation from First Harmonic. The raise takes total funding to $55 million since the company was founded in 2024.
What problem is AIUC actually solving?
Companies are deploying AI agents, software programs that take actions autonomously rather than just answer questions. The security question is whether those agents can be trusted inside a corporate system.
AIUC's answer is AIUC-1, a testing and certification standard. Before deployment, AIUC puts an agent through roughly 5,000 adversarial scenarios designed to make it misbehave. The tests look for jailbreaks (tricking the AI into ignoring its own safety rules), hallucinations (the AI confidently fabricating false information), prompt injections (hidden instructions inside a document or message that hijack what the agent does), unexpected behaviour and data leaks.
Quarterly audits follow the initial certification, keeping pace as new attack methods emerge.
Why is this getting attention now?
Rune Kvist, AIUC's founder and previously Anthropic's first product hire, put it plainly: "Most enterprises have a list of AI agents that were approved in pilots but stalled at the security review. Evidence of security and reliability is now the main bottleneck."
That matches what security teams are saying. Buying an AI tool is straightforward. Getting it through an internal risk review, without independent evidence of testing against realistic failure modes, is the part that stalls.
Seven well-known tools, including Cursor, an AI coding assistant, and Harvey, an AI legal research platform, already carry AIUC-1 certification. KPMG and UiPath are also on that list. Meaningful early traction for a company barely a year old.
Our 26 August story on who pays when AI agents act without authorisation showed courts, insurers and regulators only beginning to work out liability. A certification framework that generates a documented audit trail is one partial answer to that gap.
The new capital goes toward extending AIUC's work to frontier models, where the safety stakes are higher and independent testing infrastructure is thin. Whether a certification standard can keep pace with how quickly these models change is the real question.
Should you worry?
For employees using AI tools at work, one step is worth taking: ask your IT or security team whether the products your company uses have passed any independent safety review. If the answer is no, that's worth pushing on.
