Nuclear

The current revival of the nuclear industry almost exclusively serves the demands for AI power, raising “significant concerns about whether the risks associated with nuclear facilities and unsubstantiated, fast-tracked initiatives can be justified.”¹ Ironically, the pretense of an AI arms race with China is being used to “discard the very risk and safety thresholds established by the nuclear-arms-race era amid the threats of the Cold War.”² The government justifies this elevated risk level by arguing that accelerated adoption of AI will give the US a technological edge over its global adversaries and lead to widespread societal benefits—both claims undermined by the questionable efficacy of AI-based systems.³ Given the untested and unproven nature of AI technologies and catastrophic risks of expanding nuclear power without robust safeguards, federal policymakers should reject efforts to rapidly scale the nuclear industry to meet the demands of AI firms.

1. Sofia Guerra and Heidy Khlaaf, “Fission for Algorithms: The Undermining of Nuclear Regulation in Service of AI,” AI Now Institute, November 2025, https://ainowinstitute.org/publications/fission-for-algorithms. ↩︎
2. Ibid at 15. ↩︎
3. Ibid at 15; AI Now Institute, “Consulting the Record:AI Consistently Fails the Public,” June 3, 2025, https://ainowinstitute.org/publications/research/3-consulting-the-record-ai-consistently-fails-the-public. ↩︎

In addition to rolling back long-established safety thresholds, the government is concerningly using AI technologies to expedite regulatory processes, such as nuclear licensing and commissioning for civil and defense nuclear facilities. Nuclear licensing is a well-established process that requires nuclear operators to demonstrate that the risks arising from their lifetime operations will be adequately controlled, and to take responsibility for controlling and addressing these risks.¹ Introducing generative AI to “streamline” nuclear licensing increases the likelihood that mistakes will arise in the process. Research has consistently demonstrated the lack of accuracy of generative AI and LLMs, including a high rate of inaccurate results,² high hallucination rates,³ and a demonstrated bias in widely used LLMs toward overgeneralizing scientific conclusions.⁴ Even minute errors can compromise nuclear safety thresholds and lead to catastrophic consequences, including widespread radiation exposure.⁵ The federal government must ban the use of generative AI systems in the nuclear-energy permitting process, where safety is at stake.

1. Guerra and Khlaaf, Fission for Algorithms, 22–23. ↩︎
2. Ibid., 27; Jason Wei et al.,“Measuring Short-Form Factuality in Large Language Models,” arXiv, November 7, 2024,
   https://doi.org/10.48550/arXiv.2411.04368. ↩︎
3. Jeremy Hsu, “AI Hallucinations Are Getting Worse – and They’re Here to Stay,” New Scientist, May 9, 2025,
   https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay; Maxwell Zeff, “OpenAI’s New Reasoning AI Models Hallucinate More,” TechCrunch, April 18, 2025,
   https://techcrunch.com/2025/04/18/openais-new-reasoning-ai-models-hallucinate-more; Gyana Swain, “OpenAI Admits AI Hallucinations Are Mathematically Inevitable, Not Just Engineering Flaws,” ComputerWorld, September 18, 2025, 
   https://www.computerworld.com/article/4059383/openai-admits-ai-hallucinations-are-mathematically-inevitable-not-just-engineering-flaws.html; Guerra and Khlaaf, Fission for Algorithms, 27. ↩︎
4. Uwe Peters and Benjamin Chin-Yee, “Generalization Bias in Large Language Model Summarization of Scientific Research,” Royal Society Open
   Science 12 (March 2025): 241776, https://royalsocietypublishing.org/doi/epdf/10.1098/rsos.241776; Guerra and Khlaaf, Fission for Algorithms, 27. ↩︎
5. Ibid., 23. ↩︎

Established in 1974, the Nuclear Regulatory Commission (NRC) operated as an independent regulator, and thus was shielded from political or industry influence. Through a series of executive orders in 2025, the White House has slowly pulled back the NRC’s independence, first enabling the Office of Management and Budget to oversee the regulatory process¹ and later requiring the NRC to establish shortened timelines on nuclear licensing.² The EOs also asked the NRC to consult about radiation exposure limits with the Department of Defense and the Department of Energy, two agencies that are incentivized to expedite AI adoption and lack relevant experience with radiation exposure limits.³ Federal policymakers can reinstate the NRC’s independence to ensure nuclear safety is not compromised by political aims.

1. Ibid., 14; White House, “Ensuring Accountability for All Agencies,” February 18, 2025, https://www.whitehouse.gov/presidential-actions/2025/02/ensuring-accountability-for-all-agencies. ↩︎
2. Guerra and Khlaaf, Fission for Algorithms, 14; White House, “Ordering the Reform of the Nuclear Regulatory Commission.” ↩︎
3. Guerra and Khlaaf, Fission for Algorithms, 14. ↩︎

Risk analysis frameworks for nuclear energy date back to the earliest moments of discovery at the height of the Cold War.¹ Risk and safety thresholds have developed and changed over time, substantially influenced by catastrophic disasters and the resulting shifts in public perception. President Harry S. Truman’s decision to use atomic weapons against Japan and the 1945 Trinity test nuclear fallout, which reached forty-six states over ten days, led to the establishment of radiation dose limits and targets based on the concept of the linear no-threshold model (LNT), which determined that risk increases linearly with dose, and that no level of exposure is risk-free. This model enjoys wide-ranging international consensus and is based on long-term, detailed studies. Safety analysis also developed the “as low as reasonably achievable” (ALARA) risk principle to ensure that catastrophic risks are kept within tolerable societal limits. Despite this, the White House’s May 2025 Executive Order recommended dismantling LNT, a key safety pillar, and its consequential standard ALARA, rejecting long-standing international consensus and increasing the public’s risk of exposure to ionizing radiation.

1. Guerra and Khlaaf, Fission for Algorithms, 7–8. ↩︎