How to Validate AI Under FDA's Computer Software Assurance Guidance

September 24, 2026, 11:00 am EST

FDA's Computer Software Assurance guidance already gave teams a more efficient path to validation by encouraging a risk-based approach. AI puts that framework to its hardest test yet. When a system's output can shift with every model update, scripted test cases alone stop being sufficient, and teams are left asking how to prove AI can be trusted in a regulated process. This session breaks down how to apply CSA's risk-based principles to AI, and what that looks like in practice, using a validated Requirement Conflict Detection AI agent as a real-world example.

What you'll learn

• What changes when validation has to account for model drift, prompt sensitivity, and non-deterministic outputs
• A risk-based framework for evaluating an AI for regulated use including accuracy against intended use, level of human oversight, and reproducibility, among other dimensions
• Strategies to automate validation and reduce manual documentation effort using data from your existing tools

Who should attend

Speakers

Bailey Canter

Director of Solutions
Ketryx
Formerly Scrum Master, Amwell & Systems Engineer, Raytheon