One of the most critical tasks in licensing a nuclear reactor system is the verification and validation (V&V) processes required by the U.S. Nuclear Regulatory Commission. This step is crucial to ensure that these systems are hardened and resilient for the ensuing years of operation. The current rise of AI tool availability has been accompanied by questions about how these adaptive and dynamic systems could pass through the necessary steps from a V&V perspective. This talk will discuss the landscape of V&V tasks from software development, to safety, to human factors.

While we use V&V as a (seemingly) single concept it in fact captures two separate, but linked, efforts. First, verification is the step that determines if the system was built ‘right’ which is typically evaluated using technical audits, expert evaluations, and alignment with use case requirements. The next step is validation, which determines if the thing that was built was the ‘right’ thing. Does the tool, system, component serve its purpose in the context of the specific application space.

This talk will review these concepts and potential issues with dynamic or adaptive technologies. The focus will involve reviewing common standards, expectations, and proposed regulations in Part 53 for advanced reactors. What impact do these documents or regulations have on the AI development process and is there a path to achieving the level of consistency and assurance that NRC requires to approve these systems? If not, is this an opportunity for a new V&V process that occurs at regular intervals in perpetuity?