Leading AI Governance: Synthetic Truth Governing Generative AI in High Stakes Domains

How do you tell your board your AI is "safe" when you can't actually see inside the model making the decisions? This conversation tackles the uncomfortable middle ground between clean data and trustworthy output — the space where AI doesn't just get things wrong, it makes things up. Kelle O'Neal and Lisa Wintrich of First San Francisco Partners recap the first six months of the Leading AI Governance series before introducing "synthetic truth" — AI output that looks true but isn't. They cover the difference between fabrication and hallucination, why models are incentivized to guess confidently rather than admit uncertainty, and how the "decision steward" and "pod" concepts govern AI decisions that data governance alone can't reach. The discussion applies a consequence framework built around reversibility, defensibility, and evidence to a real-world example: a bank using AI to draft adverse action letters. It closes with audience Q&A on semantic layers, third-party vendor AI, and how "if it isn't codified, it isn't real" applies to accountability in agentic systems. Most relevant for data governance, AI governance, and risk professionals responsible for defending AI-driven decisions to leadership and regulators. Download slides: https://content.dataversity.net/rs/65... More Leading AI Governance on-demand webinars: https://www.dataversity.net/resources... Upcoming Leading AI Governance webinars: https://www.dataversity.net/events/?_... View all upcoming online events: https://www.dataversity.net/events/?_...