The International Symposium on Artificial Intelligence and Nuclear Energy in Vienna highlighted that the rapid integration of AI is shifting the nuclear supply chain from managing physical components to governing a fast-mutating parallel digital ecosystem. Global regulators and industry leaders warned that this transition introduces high-stakes vulnerabilities, such as a talent shortage, software defects and cybersecurity risks.
ATOMS FOR ALGORITHMS: Global Regulators and Tech Innovators Plan to Safeguard the AI Nuclear Supply Chain
VIENNA — While the international community remains captivated by the promise of clean nuclear infrastructure powering energy-hungry artificial intelligence data centers, a high-stakes gathering of global energy brokers and technology architects at the International Atomic Energy Agency (IAEA) revealed a more immediate challenge: the rapid and highly complex integration of AI into the global nuclear supply chain itself.
On Day 2 of the International Symposium on Artificial Intelligence and Nuclear Energy—Thursday, December 4, 2025—the focus turned directly to these vulnerabilities and opportunities during a session titled Panel 5: Nuclear Supply Chains: Current Approach and Expanded use of AI. Over the course of 90 minutes, an elite cross-section of nuclear regulators, industrial vendors, and software engineers laid out a blueprint for how the sector can capitalize on automation without losing its absolute grip on safety and security.
The consensus from the summit floor was clear: nuclear supply chains are among the most complex in the energy sector, built to maintain operations over a span of 60 to 80 years across a highly fragmented international regulatory landscape. Yet, as AI transitions from a corporate buzzword into an active utility, the industry faces an unprecedented shift: it must now govern not just the physical supply chain of steel and valves, but a parallel, fast-mutating digital supply chain.
The Digital Threat and 3D-Printed Realities
Delivering the opening keynote, Mr. Kevin Lee, Lead for Disruptive, Innovative and Emerging Technologies at the Canadian Nuclear Safety Commission (CNSC), challenged delegates to reconsider what a nuclear supply chain actually looks like. Lee, whose specialized team prepares Canada to regulate advanced technologies, revealed a thought experiment his agency conducted five years ago that is rapidly becoming a commercial reality.
“The thought experiment was way in the future, you know, 20, 30 years from now, what might the supply chain look like?” Lee explained. “And what we came up with was that there was an AI entity… who was in charge of predictive maintenance. It predicts that a pump is going to fail in two days. But that AI entity is also in charge of advanced manufacturing. So he says, ‘You know what? I’m going to print that pump.’ Pump gets printed and I say, ‘Well, you know what? I’m also in charge of installation and I have drones and robotics that I can use to install this pump.'”
While that highly automated ecosystem was initially framed as a mid-century concept, Lee told the audience, “I don’t think we’re 30 to 40 years away from that.”
This acceleration brings serious, unintended consequences. Lee observed that international groups are overwhelmingly focused on physical materials, entirely ignoring the software supply chains driving next-generation facilities like commercial fusion plants. Recalling a recent interview with a major consulting firm surveying future fusion supply chains, Lee asked them about their digital and AI supply strategy, only to be told, “Well, no, no one’s talked about that yet… that wasn’t part of what we were supposed to ask.”
Lee warned that as the industry leans on automated logistics, the threat landscape shifts entirely. Managing issues like Counterfeit, Fraudulent, and Suspect Items (CFSI) will look completely different than it did half a decade ago, especially as malicious threat actors leverage “vibe coding” to inject unauthorized code into critical networks.
Furthermore, Lee noted that deploying these technologies across a regulatory agency like the CNSC—which oversees everything from uranium mines to medical isotopes and nuclear waste management across 1,000 employees—presents a massive talent hurdle. “The question then becomes where do you get the human resource in that supply chain who understand AI and understand nuclear,” Lee said. “We’re still debating this—is it easier to get someone who understands nuclear and upskill them on AI, or are we to bring in someone who understands AI and upskill them on nuclear?”
Managing ‘Defects’ in a Fragmented Market
The complexity of the current market landscape was echoed by Dr. Andy Vickers, Vice President and Chief Technical Officer of Generative AI at Capgemini Engineering. Vickers reminded the room that the lightning-fast delivery of AI systems lacks the traditional, peer-reviewed academic rigor that the nuclear sector historically relies upon.
“New technology appears weekly almost, and often it’s non-linear… and language around topics like reasoning, trust, full AGI… they’re used without that much rigor if the truth be told, and they’re driven by the need to drive economic arguments,” Vickers stated flatly.
Vickers urged the sector to cast aside the generic label of “adopting AI,” comparing it to saying “let’s adopt maths.” Instead, he noted that the true industrial transformation lies in moving beyond text or code generation into generative biology, chemistry, and manufacturing control, alongside “Agentic AI”—autonomous digital networks capable of blurring boundaries across corporate silos and supply chains.
However, Vickers leveled a stark warning against treating “human-in-the-loop” protocols as a universal safety net. When systems become overly complex, humans naturally lose situational awareness, often trusting probabilistic models because they sound deceptively human. Vickers also targeted the popular tech euphemism for software errors.
“We need to get a little bit past the new word ‘hallucinations,'” Vickers said. “It’s rare for my boss to tell me that I’ve hallucinated. I normally have just cocked up. Yeah, I made a mistake. We need to be really talking about defects and defect management strategies.”
The Energy Paradox: Problem and Solution
Representing the global information technology layer was Mr. Marc Kamphausen, Senior Vice President of Global Customer & Partner Success Management at Oracle. Kamphausen, a nuclear engineer by trade who transitioned into the software industry, noted that the relationship between computing and nuclear power is deeply reciprocal.
“AI is also part of the problem… because we know AI consumes a lot of energy,” Kamphausen told delegates, pointing out that Oracle acts as a primary backbone for global AI infrastructure and is building super-large data centers powered directly by small modular reactors (SMRs). “So the interesting thing in this discussion is actually yes, AI is part of the problem of the energy consumption, but also potentially part of the solution to it.”
Kamphausen stated that approximately two-thirds of supply chain management clients across non-nuclear sectors are already utilizing AI. He noted that the true breakthrough for nuclear operators will come from combining machine learning with blockchain technology to rigorously authenticate and track materials globally.
To move forward safely, Kamphausen advocated for an agile corporate mindset. “We call it a crawl, a walk, a run, a fly approach,” he explained. “That means we need to start somewhere. And we may need to experiment with models where there’s less or very little risk, where there’s humans everywhere… because if we wait too long to finalize all regulations, we can’t plan the future for next year right now.”
Rebuilding Capacity and the Lexicon Crisis
Providing an industrial owner’s perspective was Dr. Nathan Flaman, Head of Global Growth at Cameco UK Limited. Flaman drew on his background managing massive, billion-dollar mining supply chains to highlight the constant operational friction between driving down costs and mitigating systemic risk.
Flaman observed that the historical “boom-and-bust” cycles of the nuclear industry have left the global supply chain deeply cautious, hesitant to invest heavily in expanding capacity without guaranteed market certainties. This is further aggravated by a highly fragmented market where suppliers are forced to try and guess which reactor designs will win final investment decisions.
However, Flaman noted that advanced data models are beginning to allow operators like Cameco and Westinghouse to closely integrate with vendors to safely share data across national borders. “If something’s not available in a particular region and it’s needed somewhere else, we can actually have some further resilience there which is enabled when you have a reasonable level of uniformity,” Flaman noted.
Yet, this vision of seamless international data sharing faces a fundamental, almost absurd obstacle: the nuclear industry doesn’t speak the same language.
To illustrate this point, Kevin Lee recalled a past conversation with a seasoned industry veteran who told him, “In Canada, you have Bruce Power, you have OPG. They call a widget different things. It’s the same widget, but they call them different things between the two operators… OPG has two plants… They call different things widgets differently between those two plants. There is no standardization of language of lexicon in the nuclear industry.”
Lee pointed out that the global introduction of standardized SMR designs offers a rare, historic opportunity to use AI to build and enforce a uniform international lexicon from the ground up.
International Cooperation: No One Walks Alone
The panel concluded with an emphatic demand for cross-border cooperation. Lee highlighted a series of active international partnerships, including a trilateral innovation white paper published in September 2024 by the US Nuclear Regulatory Commission (NRC), the UK Office for Nuclear Regulation (ONR), and Canada’s CNSC.
Furthermore, this pilot program has expanded into a formal project under the Nuclear Energy Agency (NEA), uniting seven global regulators to run collaborative, international “regulatory laboratories” (Reg Labs) focused on near-term AI deployment, predictive maintenance safety, and cybersecurity. Lee also proposed that the IAEA’s Innovation to Support Operating Plants (ISOP) network immediately establish a fifth dedicated working group explicitly focused on AI nuclear supply chains.
Summarizing the path forward, Andy Vickers reminded delegates that navigating this technology requires a level of humility from all global actors.
“This is changing the way we do our work,” Vickers concluded. “No one owns the truth in this space. No one owns the truth… My one final recommendation, dare I say it, for everybody is go and play. Play with the technology. It’s at your fingertips now… It’s the best way to start to understand the art of the possible.”
About the Author
