AIUC Raises $40M to Make Enterprise AI Agent Deployment Auditable
Most enterprises have exactly this problem: a growing list of AI agents that cleared the pilot stage and then stalled. Not because the agents were not capable enough. Because no one could tell the security team what they would do in production.
On September 15, 2026, TechCrunch reported that Artificial Intelligence Underwriting Company (AIUC) raised $40 million in a Series A led by Ribbit Capital. The round brings AIUC’s total funding to $55 million and is the clearest signal yet that the enterprise AI industry is converging on a formal certification layer for agents, modeled on the same compliance infrastructure that governs software security and financial auditing.
The Bottleneck Is Not Capability Anymore
Rune Kvist, AIUC co-founder and Anthropic’s first product hire, has a line he repeats to every prospective customer: “Banks, hospitals, governments and militaries no longer decline to deploy AI because a model isn’t smart enough. 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.”
The result is a widening gap between AI capability and AI deployment. By most practitioner estimates, 95 percent of enterprise AI pilots fail to reach production, and the primary reason is not technical performance. It is that security and compliance teams lack an objective, independent standard for evaluating what an agent will do in a live environment.
This problem gets worse as agents grow more capable. A simple chatbot runs in a constrained text interface. A modern AI agent accesses databases, executes code, calls APIs, sends email, and can take actions that are difficult or impossible to reverse. The attack surface is categorically different, and it changes every few months as models improve.
What AIUC Built: A SOC 2 for AI Agents
AIUC’s core product is the AIUC-1 certification standard, developed alongside a consortium of approximately 250 security and risk leaders from Fortune 1000 enterprises. The consortium meets monthly. Its input shapes the test scenarios. The goal is to ensure the standard tracks what actual enterprise buyers need to know before approving a deployment, not what the vendors building agents want to be tested on.
The certification process runs an agent through roughly 5,000 adversarial test scenarios spanning six risk domains:
| Risk Domain | What AIUC-1 Tests |
|---|---|
| Data and Privacy | Unauthorized data access, leakage, retention violations |
| Security | Prompt injection, jailbreaks, privilege escalation |
| Safety | Harmful or off-policy output |
| Reliability | Hallucinations, task completion accuracy, failure cascades |
| Accountability | Audit trail completeness, explainability of actions |
| Society | Bias, discriminatory output, downstream societal harm |
Tests are modeled on documented real-world AI failures, not theoretical edge cases. The process produces a detailed audit report showing where an agent passes, where it has known failure modes, and what conditions increase risk. Agents must pass quarterly recertification tests to maintain their certification as threat vectors and model capabilities evolve.
AIUC co-founder Rajiv Dattani, the former COO of AI safety research organization METR, described the goal plainly: “Here’s where it passes and where you can trust it. And here’s where there’s concerns. You should be aware of those references before you make the decision to buy.”
The analogy to SOC 2 is deliberate. SOC 2 did not make software secure. It gave procurement teams a common framework for evaluating security posture, which normalized the conversation and accelerated purchasing. AIUC-1 is attempting the same translation for AI agent risk.
The Insurance Layer Is the Real Differentiator
Certification standards without consequences are advisory documents. AIUC went further.
The company partnered with Lloyd’s of London to provide insurance coverage for certified AI agents. When ElevenLabs received the first AIUC-1-backed policy, it carried $50 million in coverage for risks including hallucination-driven loss, data leakage, and faulty tool actions. The “underwriting” in AIUC’s name is not a metaphor. The company is literally putting capital behind its certifications.
This changes the dynamic for enterprise procurement. A vendor with AIUC-1 certification plus a Lloyd’s-backed policy is not just asserting that their agent is safe. They are accepting financial liability for specific failure modes. For regulated industries, that is a fundamentally different conversation than reviewing a security whitepaper.
Micky Malka, founder of Ribbit Capital and lead investor in the Series A, framed the investment thesis around this model: “We have spent over a decade backing companies in financial services where trust is the most important metric of success. AI is on that same path, and it is moving faster than the systems companies use to evaluate it.”
Ribbit’s portfolio has historically concentrated on fintech infrastructure: the category of companies whose value comes from being the trusted intermediary between counterparties in high-stakes transactions. AIUC is that model applied to AI agent deployment.
Who Is Already Certified
AIUC-1 certification is not aspirational. As of September 2026, SecurityWeek reports that the following agents have been certified or are actively in the certification process: Cursor, ElevenLabs, Fin by Intercom, Harvey, KPMG, Lovable, and UiPath.
The KPMG milestone is significant. KPMG became the first Big Four professional services firm to achieve AIUC-1 certification in August 2026. Big Four firms operate under some of the strictest client data obligations in professional services. Their willingness to seek external certification, rather than rely on internal audits, signals how the standard is evolving from a vendor-facing badge to a client-facing commitment. The AIUC-1 standard is also listed in the CSA STAR Registry, allowing certified companies to display a trustmark in a vendor assessment context that enterprise procurement teams already use.
The Founding Team and the METR Connection
The founding story matters for enterprise buyers evaluating AIUC’s credibility.
Rune Kvist joined Anthropic as its first product hire. He left to co-found AIUC because he saw that capability improvements were creating deployment risk, not reducing it. “AI is getting smarter at an increasingly rapid rate. The surprising thing about AI is that it becomes harder to adopt and harder to control as AI gets smarter, not easier,” Kvist told TechCrunch.
Rajiv Dattani served as COO of METR from 2024 to 2025 and remains a METR board member. METR is the safety research organization that conducts autonomous-capability evaluations for frontier labs including OpenAI, and was one of the independent researchers OpenAI used to investigate its Hugging Face incident. Dattani’s background brings the methodology of frontier AI safety evaluation to the enterprise deployment context.
The seed round included Nat Friedman (NFDG), Emergence Capital, Terrain, and Anthropic co-founder Ben Mann, an investor lineup that reflects the team’s deep roots in the safety-adjacent part of the AI industry.
The Broader Context: From Frontier Safety to Enterprise Deployment
The AIUC funding arrived during the same week that Anthropic CEO Dario Amodei published an essay calling on the AI industry to collectively slow frontier model development to avoid catastrophic outcomes. Amodei’s essay specifically floated requiring frontier labs to embed third-party evaluators to observe and verify safety from the inside, naming METR as one possible model.
Amodei’s call for embedded evaluators at the frontier level and AIUC’s certification model at the enterprise deployment level are addressing the same underlying problem from different angles: the gap between AI capability and the governance infrastructure needed to deploy it responsibly at scale.
For enterprise AI and GTM leaders, the gap is immediate and practical. There is a category of agents that could be in production today, generating ROI, but which are sitting in a security review queue because no one has a standard framework for evaluating them. AIUC-1 is the most developed attempt to build that framework as of September 2026.
What Enterprise Leaders Should Do Next
The AIUC Series A is a directional signal, not just a funding story. Three actions are practical today:
-
Audit your pilot backlog. Identify every AI agent in your environment that completed a pilot but has not been approved for production. The blocking issue in most cases is security review. Categorize which risks are being cited and whether a third-party certification against AIUC-1 would address them.
-
Require certification in procurement. If your organization is evaluating AI agent vendors, add AIUC-1 certification (or an equivalent independent audit) to your vendor security questionnaire. Vendors who cannot provide independent certification of their agent’s behavior should require more scrutiny, not less.
-
Review your insurance exposure. Most enterprise policies were written before AI agents became a deployment-scale reality. Talk to your broker about whether your current coverage addresses hallucination-driven financial loss, data exfiltration by an autonomous agent, or liability for actions taken by a third-party certified agent. The Lloyd’s model AIUC established with ElevenLabs is a template for how this coverage can work.
If you are evaluating where AI agents fit in your enterprise architecture, Enera can help. For a broader view of how the AI agent security category is forming, see our analysis of Zenity’s $125M raise and the AIR Security $50M agent firewall.
The security review is not going away. The question is whether your organization has the infrastructure to clear it at the pace your AI roadmap requires.