Excerpted from this report:

The Opportunity: Two Moments, One Connection

Two things happened at roughly the same time, and few experts have connected them.

The first is regulatory. Congress, through the Nuclear Energy Innovation and Modernization Act (NEIMA), directed the NRC to establish a technology-inclusive framework and to increase its use of risk-informed, performance-based licensing evaluation techniques. The NRC delivered 10 CFR Part 53, effective April 29, 2026, more than a year ahead of the statutory deadline of December 31, 2027, and has since proposed a broader modernization package that carries the same risk-informed, performance-based logic into Parts 50 and 52. The second is technical: agentic AI has matured to the point where it can carry out the labor a formal performance-based licensing approach demands. This report is about what happens when the two are put together.

By agentic AI we mean systems that can autonomously plan and execute multi-step work, decomposing a goal, selecting tools, iterating on outputs, and maintaining a structured record of their reasoning, rather than simply responding to a single prompt.

This is a policy argument, not a persuasion piece. We are not trying to convince anyone that Part 53 is superior to Parts 50 and 52, or that performance-based licensing should displace probabilistic methods. We are pointing to a connection between a regulatory moment and a technical one, and showing where it leads. The connection matters now because the performance-based reading of Part 53 is easy to lose. Read through the habits built under Parts 50 and 52, the new framework can look like another procedural path, and the flexibility long sought by industry disappears. That pattern is not hypothetical: it appeared with 50.69, the Reactor Oversight Process, aging management, post-Fukushima FLEX upgrades, and the Licensing Modernization Project. Each offered flexibility; each was slower to take hold than its benefits warranted, and the reason was not technical.

Our thesis is that Part 53 is being read the wrong way. Applicants seem to approach it as another procedural path to a license, evaluating it by which processes it requires them to execute. Read that way, Part 53 looks like Parts 50 and 52 with unfamiliar section numbers, and the rational move is to stay with the framework you know. Read on its own terms, Part 53 is something different: a framework that relies on structured decision-making rather than a prescribed set of processes. That distinction is the difference between demonstrating compliance and exercising design freedom, and it carries direct cost and schedule consequences for anyone deploying a novel reactor.

The methodology was always sound. What was missing was a way to carry its labor at the speed a project demands. That is what has now changed.

 

References

RTS, Nuclear AI as the Key to Part 53: Using AI Agents to Implement Riks-Informed, Performance-Based Licensing, September 2026, by Sam Rieck and N. Prasad Kadambi.