Brand Safety Videos

 AI Data and Training
with Katrin Ribant

BSI’s Louis Jones recently sat down with Katrin Ribant of Ask-Y to explore the often-overlooked intersection between brand safety and how AI systems are trained.

Ribant, former Technology Lead for HAVAS Digital and co-founder of Datorama (sold to Salesforce), demystifies AI training through her “Floofies”, illustrating how large language models evolve and gradually shape themselves into coherent systems. While deeply technical, this process has direct implications for advertisers and brands. Hallucinations, bias, and risk are not random outcomes, but learned behaviors. Effective data analytics and insight depend on how context is established, preserved, and governed by both human teams and the AI tools they deploy.

They explore why safe AI begins with sound training and transparent design; what happens when AI development is no longer a black box; and how clearer visibility into analytics, memory, and decision-making strengthens trust and accountability. From early-stage training concepts to agentic analytics in practice, the discussion offers an inside look at what AI truly is before it ever acts, drawing on insights from Ask-Y’s first year of building and structuring its analytical systems.

In an ecosystem racing toward automation, this conversation makes the case that how AI learns may be the most important brand safety perspective of all.