Nuclear Regulatory Commission in ‘testing’ phase with Azure OpenAI

The Nuclear Regulatory Commission has started evaluating a potential application of the Microsoft Azure OpenAI tool that would leverage its data repository, the agency said in a call with reporters Monday. 

Basia Sall, the chief data officer and director of the NRC’s Data, Information Management and Enterprise Governance division, said that the Azure OpenAI effort is still in its early stages, with pricing and security details still under review. Agency staffers have“just really started getting some data” out of the system, Sall said, adding that there haven’t been any internal briefings yet on the initial findings but there will be “more to come.” The agency is also looking to release a gap analysis and other documents throughout September.

“We’re going to be using that Microsoft tool to help us work with our large data set to pull out generative AI and use it for knowledge management, at least in the beginning,” Sall said. “But that’s our test case we’re doing right now. We’re doing some work on being able to pull some data out, validating that we’re getting the responses we want.”

The OpenAI through Azure application didn’t need existing FedRAMP approvals, since the Microsoft program was already FedRAMP-certified, Sall said. She noted that there was no additional action needed, which was “a huge benefit to us as we know that [FedRAMP certification] gives us that additional level of security and confidence in the oversight of especially our high-value assets.”

Monday’s update makes the NRC the latest federal agency to experiment with large language model applications, and specifically software designed by OpenAI. The news also comes amid the Biden administration’s ongoing effort encouraging the federal government to adopt AI tools. While the executive order doesn’t explicitly apply to an independent federal agency like the NRC, staff appear to be following the spirit of the White House’s AI objectives anyway.

Late last month, Anna Makanju, the vice president of global affairs at OpenAI, told FedScoop that USAID was the company’s first federal customer for its ChatGPT Enterprise system. NASA has also been experimenting with the use of OpenAI through Microsoft Azure. Last month, Azure OpenAI received FedRAMP High authorization status.

The update is just the latest in the NRC’s broader artificial intelligence-related efforts. In April, the agency released an informational policy document detailing its approach to AI. That document outlined dozens of potential use cases for the agency and two possible next steps, including creating an enterprise AI strategy and investing in AI foundation tools. 

Just last week, the agency released a document in concert with its Canadian and U.K. nuclear regulatory counterparts on the use of AI within the nuclear sector. Together, the North American and British agencies highlighted the need for AI system boundaries, modularizing AI applications and nonstop monitoring of systems. 

Officials also announced Monday that the NRC is expecting to release a gap analysis concerning the agency’s next steps and readiness for AI. Victor Hall, the deputy director of the Division of Systems Analysis in the Office of Nuclear Regulatory Research, said the agency had experts weigh in on the report.

“I think we have done all the prep work to say … we do have the expertise in-house to be ready for AI in the nuclear industry, recognizing the industry as its own unique beast,” Hall said. “And that’s something I’m pretty proud of.” 

The document is expected at the end of the month.