Voice AI Needs Its SOC2 Moment

What does it actually take to prove a voice AI agent is safe? In this episode, Coval CEO Brooke Hopkins sits down with Ian Spaulding (CEO of LRQA) and Mike Goh (co-founder of Krew) to break down the first independent assurance audit of a voice AI agent — and what it revealed about AI compliance, risk tolerance, and the future of regulated AI deployment. Krew builds AI voice agents for loan servicing teams at major U.S. financial institutions. LRQA — a 250-year-old global certification and auditing body — conducted an independent limited-assurance review of Krew's AI governance framework. Coval ran thousands of simulations to evaluate how the voice agent behaves across a wide range of real-world consumer scenarios, including edge cases, interruptions, and off-script callers. What we cover: Why voice AI compliance can't be a checklist — these systems are non-deterministic How to think about AI risk in probabilistic terms: error rates per million calls, not pass/fail Why regulated industries (financial services, healthcare) are where AI assurance standards will be defined first The SOC 2 analogy: how voluntary compliance frameworks become the industry baseline What regulators are already asking about voice AI agents behind closed doors Why combining independent governance audits with large-scale simulation gives enterprises the complete picture Where the voice AI assurance market is heading in the next 12–24 months Key insight: Most people treat AI compliance like a traditional checklist audit. But voice AI agents are non-deterministic — the same input doesn't always produce the same output. The right question isn't "did it pass?" It's "what is the failure rate per million calls, and what level of risk is acceptable for each failure type?" This is the model Waymo used for self-driving cars, and it's the model voice AI needs now. Chapters: 00:00 Introduction & guest backgrounds 03:00 What inspired the first voice AI agent audit 07:00 How LRQA approaches AI governance assurance 10:00 Why traditional audits aren't enough for non-deterministic AI 13:00 The SOC 2 analogy for voice AI compliance 16:00 What financial services regulators are asking about AI 22:00 Testing voice agents at scale — thinking in probabilities, not pass/fail 26:00 The Waymo parallel: how self-driving car safety standards apply to voice AI 31:00 Where voice AI regulation and assurance is headed in 12–24 months 35:00 Closing thoughts Links: Learn more about Coval: https://coval.ai Read the LRQA x Krew x Coval pilot: https://www.lrqa.com/en/latest-news/l... #voiceai #aicompliance #aigovernance #artificialintelligence #aiagents #responsibleai #airegulations #llm #enterpriseai