Diablo Canyon, the last operational nuclear power plant in California, has faced numerous challenges and threats of closure over the past decade. Currently, it is scheduled to begin a protracted decommissioning process starting in 2029. Despite its precarious situation, the San Luis Obispo facility recently received a significant upgrade in computing technology with the installation of eight NVIDIA H100 processors, renowned as some of the most powerful graphics processing units available globally. This advanced hardware is intended to support a new artificial intelligence initiative aimed specifically at enhancing operations within the nuclear energy sector.

Pacific Gas and Electric (PG&E), the operator of Diablo Canyon, announced a collaborative agreement with Atomic Canyon, a startup specializing in artificial intelligence based in the same region. This partnership was highlighted in a press release as a groundbreaking achievement, being termed “the first on-site generative AI deployment at a nuclear power plant in the United States.” This innovative move underscores the potential for integrating AI technology into traditional energy systems.

The Neutron Enterprise, the AI tool developed through this partnership, is designed to assist plant personnel in navigating a vast array of technical documents and regulations — encompassing millions of pages of detailed content from the Nuclear Regulatory Commission that span decades. While currently focused on enhancing the efficiency of daily operations and regulatory compliance, the existence of Neutron Enterprise hints at broader future applications of AI technology at Diablo Canyon and potentially other nuclear facilities. This prospect has prompted discussions among lawmakers and AI experts regarding the necessity for robust safety protocols and regulatory frameworks.

PG&E is implementing the document retrieval service in phases, with the installation of NVIDIA chips marking one of the initial steps in the collaboration with Atomic Canyon. PG&E anticipates a “full deployment” of Neutron Enterprise by the third quarter of this year, as stated by Maureen Zawalick, the company’s vice president of business and technical services. Once fully operational, Neutron Enterprise is expected to function as a data-mining “copilot” to assist in searching and summarizing specific instructions and reports pertinent to Diablo Canyon’s operations, though it is explicitly not intended to replace human decision-making.

“We estimate that our team spends approximately 15,000 hours annually combing through various databases, records, and procedural documents,” Zawalick remarked. “The implementation of Neutron Enterprise is projected to significantly reduce that time commitment, allowing our staff to focus on more critical tasks.” This reduction in manual labor not only enhances operational efficiency but also enables personnel to allocate their time to more pressing responsibilities.

Trey Lauderdale, the co-founder and chief executive of Atomic Canyon, articulated his straightforward objective for Neutron Enterprise: to empower employees at Diablo Canyon to retrieve relevant information swiftly and effectively. “You can document this: the AI expert in nuclear energy firmly states that I absolutely do not want AI managing our nuclear power plant at this moment,” Lauderdale emphasized. This cautious approach reflects a broader sentiment within the industry, recognizing the importance of gradual integration of AI technologies in such sensitive environments.

The phrase “at this moment” is particularly significant. Both PG&E and Atomic Canyon share a consensus on maintaining a cautious and limited application of AI for the foreseeable future, while leaving the door open for expanded utilization of AI tools at the plant as conditions evolve. Lauderdale revealed that his company is also engaging with other nuclear facilities and entities interested in developing small modular reactors to explore potential integrations of their technology, highlighting a growing interest in applying AI within the nuclear energy domain.

Ongoing concerns linger regarding the adequacy of existing safeguards to manage the risks associated with the convergence of artificial intelligence and nuclear technology. The Nuclear Regulatory Commission has been investigating the implications of AI in nuclear operations for several years, yet it remains uncertain whether this issue will continue to receive attention under the current administration. Shortly after taking office, former President Trump rescinded a previous executive order from the Biden administration that outlined regulatory objectives for AI, claiming that such measures could hinder American innovation. In the meantime, Atomic Canyon is proactively keeping the Nuclear Regulatory Commission updated on its initiatives and plans.

Tamara Kneese, the director of the Climate, Technology, and Justice program at the nonprofit Data & Society, acknowledged that a narrowly defined document retrieval service like Neutron Enterprise could boost efficiency. However, she expressed skepticism regarding the notion that generative AI could be limited to just one specific function within a nuclear facility, emphasizing the need for ongoing scrutiny of PG&E’s implementation of AI technology in such a critical context.

For these reasons, Democratic Assemblymember Dawn Addis, who represents San Luis Obispo, has raised concerns about the recent advancements at Diablo Canyon. “I have numerous unresolved questions regarding the safety, oversight, and implications for employment associated with the implementation of AI at Diablo,” Addis stated. “Previously, I have supported initiatives aimed at regulating AI and preventing job displacement through automation. It is imperative that we establish necessary safeguards, especially when deploying AI technologies in sensitive environments like Diablo Canyon.”

Understanding the Journey of AI Integration in San Luis Obispo

Trey Lauderdale’s journey into the intersection of artificial intelligence and nuclear energy began with his founding of a healthcare software company named Voalte, which facilitated communication among hospital staff through iPhones, thereby reducing dependence on loudspeaker paging systems and desktop computers. Back in 2008, Lauderdale encountered skepticism from hospital personnel regarding his innovative proposal. He often draws parallels between that initial experience and the current resistance he faces with Atomic Canyon’s AI solutions.

In 2021, Lauderdale relocated to San Luis Obispo to be closer to his wife’s family in Northern California, only to discover the proximity of Diablo Canyon to his new home. Through interactions with local workers from Diablo Canyon, he gained insights into the complexities of nuclear energy, which inspired him to embark on his next entrepreneurial venture.

Atomic Canyon commenced operations in 2023 with a mission to download approximately 53 million pages of publicly accessible documents from the Nuclear Regulatory Commission, which encompass the entirety of the nation’s nuclear energy infrastructure and can be accessed via a database known as ADAMS. This ambitious endeavor began in January 2024, following Lauderdale’s proactive communication with the Nuclear Regulatory Commission to inform them of Atomic Canyon’s intentions: “I contacted the [the commission] to clarify that I am Trey Lauderdale, an American entrepreneur, and we are initiating an AI project in the nuclear domain, ensuring that our activities are not misconstrued as malicious attempts by foreign entities.”

Lauderdale noted that the commission was supportive of Atomic Canyon’s initiatives. After successfully downloading the necessary data, Atomic Canyon collaborated with the Department of Energy’s Oak Ridge National Laboratory in Tennessee to commence research and development efforts. This laboratory is home to the Frontier supercomputer, which was the fastest computing system in the world when it was launched two years ago. Atomic Canyon utilized the capabilities of Frontier to develop an AI model that employs “sentence-embedding models,” which are adept at comprehending complex nuclear terminology while minimizing the likelihood of incorrect or fabricated responses.

In the spring of 2024, Lauderdale and representatives from PG&E began formal discussions to explore how Atomic Canyon could contribute to operations at Diablo Canyon. Subsequently, PG&E invited Atomic Canyon staff to the nuclear facility for a hands-on experience, where they spent several weeks shadowing employees and identifying operational inefficiencies that AI could potentially address.

By September 2024, Atomic Canyon announced the successful testing of its AI system, known as “FERMI.” These open-source models collectively form the foundation of the Neutron Enterprise software. Just a few months later, in November, the collaboration with PG&E was officially announced as the first instance of AI technology being applied in the nuclear energy sector.

Exploring the Functionality of Neutron Enterprise

PG&E integrated NVIDIA hardware at Diablo Canyon to facilitate the operation of the FERMI system. Both Zawalick and Lauderdale confirmed that the Neutron Enterprise software is being installed without cloud connectivity to ensure that sensitive, internal documents remain secure and confined to the facility. Zawalick assured that their data storage practices comply with all Nuclear Regulatory Commission and Department of Energy regulations regarding nuclear information, and these practices will undergo continuous testing and evaluation.

Currently, initial users of Neutron Enterprise are utilizing the software to access publicly available regulatory data. PG&E and Atomic Canyon aspire to commence the next phase of Neutron Enterprise’s implementation by the third quarter of 2025, at which point additional on-site employees will have access to the service, allowing it to search for and summarize internal documents through the use of optical character recognition (OCR) technology, which enhances the indexing of documents, and retrieval-augmented generation (RAG), which supports more flexible querying.

According to Lauderdale, utilizing artificial intelligence to expedite document searches carries minimal risk. If the AI system fails to locate the required information, users can revert to traditional methods of searching through multiple on-site databases or manually retrieving paper files. This dual approach ensures that critical information remains accessible regardless of the AI’s performance.

Neutron Enterprise also generates concise summaries of documents during user searches, although there is a possibility that these summaries may contain inaccuracies. Nevertheless, they will not impact the actual content or instructions contained within the original documents reviewed by workers, preserving the integrity of the information.

CalMatters engaged with several state lawmakers, particularly those representing the Diablo Canyon area, to gather their perspectives on Atomic Canyon’s innovative partnership with PG&E. The overall sentiment was positive, albeit with a recognition of Neutron Enterprise’s currently limited capabilities.

Democratic Senator Henry Stern from Malibu, a member of the Senate Energy Committee, expressed cautious optimism, stating that he is “reluctant to dismiss AI tools that can enhance grid management,” provided that stringent safety protocols are adhered to. Senator John Laird, who represents San Luis Obispo, took a balanced view, asserting, “As AI integration expands, so does its energy demand… It is crucial to balance technological advancement with public safety, environmental responsibility, and regulatory oversight in shaping AI’s role in the future of our state’s energy landscape.” Similarly, Senator Scott Wiener from San Francisco, whose ambitious AI safety legislation was recently vetoed by the governor, echoed his Democratic colleagues’ sentiments, affirming that “if AI can enhance the operational efficiencies at Diablo Canyon, that is indeed a positive development.”

Of the five members of the San Luis Obispo County Board of Supervisors, three provided feedback on the matter. Supervisor Bruce Gibson remarked that “utilizing AI to streamline access to and organization of necessary information in this context is sensible,” emphasizing the importance of transparency and regular updates from PG&E. Supervisor Heather Moreno noted that it is beneficial that PG&E will leverage a “supercharged” search engine, adding that since it will not be employed for operational decisions, this seems to be a prudent initial step in the application of AI at Diablo Canyon. Supervisor Dawn Ortiz-Legg, a former PG&E employee, expressed her “encouragement” regarding Diablo Canyon’s collaboration with Atomic Canyon in managing the vast quantities of data generated from extensive audits and reports.

Addressing Regulatory Concerns Surrounding AI Technology

Despite the seemingly benign use of AI at Diablo Canyon at this stage, there are overarching concerns regarding the potential future applications of the technology at this and other facilities. “It is essential to exercise caution when discussing broader AI decision-making capabilities,” Wiener stated. “This underscores the critical need to enhance governmental capacity to establish standards governing the use of AI in sensitive contexts such as nuclear power plants.”

In November 2024, Nuclear Regulatory Commission Inspector General Robert J. Feitel echoed similar concerns, identifying “planning for and assessing the impact of artificial intelligence and machine learning on nuclear safety and security” as one of the nine significant challenges facing the agency. A commission-sponsored report published the previous month by the Southwest Research Institute examined AI-related regulatory gaps within the nuclear energy sector, revealing fewer than 100 identified gaps while also noting the absence of practical AI standards from external sources that could help mitigate these shortcomings. The report advocated for the development of AI-specific guidelines.

It appears that Atomic Canyon and PG&E are voluntarily maintaining communication with the Nuclear Regulatory Commission regarding their activities. “I wouldn’t claim we have an official relationship with the NRC, but we ensure they are briefed on our progress. Given our status as newcomers to the nuclear sector, surprises are not welcome,” Lauderdale remarked. He believes that the nuclear energy sector’s inherently cautious approach will serve as a natural safeguard against overly intrusive or hazardous AI integrations, although he acknowledged the necessity for regulatory measures as applications evolve to introduce potential risks.

When CalMatters first spoke with PG&E’s Zawalick in December, she mentioned having recently engaged with the Nuclear Regulatory Commission’s AI working group, serving as an advisory committee. Since that time, she has not had any further discussions with the commission regarding AI regulations, as she conveyed to CalMatters recently.

Furthermore, the Diablo Canyon Independent Safety Committee, a state-appointed entity responsible for inspecting the nuclear facility and providing operational recommendations, first learned about PG&E’s partnership with Atomic Canyon through media reports, according to the committee’s legal counsel, Bob Rathie. In December 2024 and January 2025, a committee representative participated in two fact-finding visits concerning Neutron Enterprise, engaging with PG&E staff to better understand the software. The committee concluded from their observations that the use of artificial intelligence at Diablo Canyon is “positive,” and they currently harbor no safety concerns.

Anticipating the Future of AI in Nuclear Energy

During a conversation with CalMatters, Lauderdale was traveling to another nuclear facility, the identity of which he could not disclose. He indicated that Atomic Canyon is “in discussions” with “numerous other nuclear organizations,” with plans for some “exciting announcements” expected later this year. Through its collaboration with Diablo Canyon, he aims to establish a proof of concept for existing nuclear facilities, as well as for entities interested in constructing or recommissioning nuclear facilities. Lauderdale expressed hope that Diablo Canyon’s operational lifespan could be extended beyond the current decommissioning timeline, asserting that if that is not feasible, his software could still play a vital role in the decommissioning process.

“As we build trust in our product and expand our capabilities, we will gradually select additional non-risky applications to implement, and we will continuously strive to create additional value using this innovative technology,” he stated, outlining his vision for the future.

Addressing concerns regarding whether the implementation of AI at Diablo Canyon has been subject to adequate oversight, Lauderdale reiterated that his startup’s product does not hold a significant operational role within the facility.

“I consider our company to be at the forefront of AI deployment within the nuclear sector,” he asserted, providing a forward-looking assessment that leaves a slightly open door: “and I believe we will not see AI actively managing nuclear power plants for a considerable time.”

This article was originally published on The Markup and was republished under the Creative Commons Attribution-NonCommercial-NoDerivatives license.