Making Safe AI Profitable: Standards, Audits, and Insurance | Sam Stowers (AIUC)

Sam Stowers (AIUC) on using insurance-style profit incentives and breadth-first evaluations to make AI systems measurably safer to deploy. In this FAR.AI seminar, Sam Stowers of the Artificial Intelligence Underwriting Company argues that neither regulation nor voluntary self-regulation will make AI safe fast enough, so the missing lever is economic. He draws on how fire, electricity, and automobiles were each tamed by the same three-part system, a safety standard, independent audits, and insurance that prices risk accurately, and shows how the same combination can reward AI developers for building safer products. He then walks through AIUC's breadth-first approach to evaluations: building a comprehensive, continuously updated taxonomy of AI risks drawn from real incidents, red-teaming labs, and frameworks like the EU AI Act, and using it to underwrite and price coverage. Two lightning talks follow, one on the challenges of evaluating voice and multimodal agents, and one on using self-improvement loops to sharpen attack prompts and scoring accuracy. 0:00 Introduction: Sam Stowers and AIUC 1:07 Why the incentives for safe AI are broken 1:52 Historical analogs: fire, electricity, and cars 5:18 Applying standards, audits, and insurance to AI 5:49 How it works in practice: enterprises and AI companies 6:58 Building the risk taxonomy 11:02 Breadth-first evaluations: is the risk set complete? 15:11 Underwriting and pricing AI insurance 17:58 Where this fits, and where it doesn't 19:18 Lightning talk: evaluating voice agents 27:42 How voice and emotion change evaluation 31:48 Lightning talk: self-improving attacks and scoring 41:27 Takeaways and Q&A More AI safety research: https://far.ai FAR.AI is a research nonprofit working to ensure the safe development of advanced AI.