When US President Trump signed an executive order in May 2025 seeking to streamline the adoption of nuclear energy by directing federal agencies to reconsider whether radiation protection standards have been unnecessarily strict, he reignited a debate that has smoldered in radiation science for decades. At the heart of the controversy is the linear nonthreshold (LNT) model—the idea that any amount of radiation, no matter how small, has damaging biological effects. Resolving whether LNT is a reasonable precaution or a costly misapplication of incomplete science will require not just better arguments, but better data.
The science underlying low-dose risk (less than 100 milligray, a measure of how much radiation is absorbed by living tissue) is incomplete. Decades of epidemiology and biology have neither confirmed LNT nor established a universally agreed threshold dose below which there is essentially no risk. Given the uncertainties, some experts advocate that doses should be as low as reasonably achievable (ALARA). There is also a hypothesis called hormesis, which holds that very low doses of radiation might be beneficial by stimulating cellular repair mechanisms. These are not purely scientific questions. How society weighs uncertain risks against economic and energy imperatives, whether in nuclear power, medical imaging, or occupational safety, involves value judgments that data alone cannot fully resolve.
Answering these questions will still require more research on several fronts. Here, experts do not agree on what that evidence will ultimately show. Some (including E.A.C.) hold that although LNT is not an established biological truth, it is a defensible regulatory tool—adding, however, that ALARA has been applied inconsistently, at times generating financial costs, psychological harm, and forgone societal benefits without proportionate protective benefit. Others (including B.A.U.) maintain that evidence for harm at low doses is lacking and that the real-world consequences of LNT-based regulation, from unnecessary abortions following the Chernobyl nuclear disaster in 1986 to evacuation deaths after the Fukushima nuclear accident in 2011, demonstrate that treating a contested model as settled science carries measurable human cost.
To gain more clarity, some scientists believe that completing the Million Person Study, the largest cohort of occupational radiation workers, is one necessary step. Preliminary findings suggest a detectable risk at occupational dose levels, but the statistical power of studies on the effects at the lowest-dose ranges—the ones that would affect policy—remains limited. Improving those data will require more efforts to follow up on participants in the study and ensure that the registries that collect exposure data from radiation workers are using uniform techniques.
But epidemiology alone cannot resolve the question that is central to the LNT debate. For example, cancer risks from low doses of radiation can be confused with other causes, such as lifestyle and random environmental factors. Rather than measuring who gets cancer, it would be better to look at the blood and tissues of exposed people over time for molecular biomarkers of cancer to understand whether low-dose radiation is as harmful as LNT assumes.
There is now an unprecedented breadth of data on radiation effects from occupational workers, patients, survivors of nuclear accidents, and cellular and molecular experiments. However, these data have been analyzed separately rather than integrating them, and often under the assumption of a particular model. Some argue that past reviews have been systematically biased, excluding studies that challenge LNT while favoring those that support it. One way to avoid preconceptions that shape analyses would be to use modern machine learning methods without imposing a priori assumptions about linearity, threshold, or hormesis. This would remove bias and could help identify dose-response patterns that conventional models may obscure.
A 2022 report by the US National Academies outlined 11 research priorities across the fields of epidemiology, biology, and data infrastructure. That agenda has stalled, not for lack of scientific merit, but because federal agencies judged the findings unlikely to move regulatory needles. President Trump’s executive order forces scientists and regulators to consider whether that calculus has now changed.
Generating better data is necessary but not sufficient. Future expert reviews must represent the breadth of scientific opinion and explicitly evaluate competing dose-response models, from LNT to hormesis. LNT-based policies have shaped radioactive cleanup standards, energy choices, and public responses to radiological accidents, carrying measurable human and economic costs of their own. The credibility of future findings will depend as much on how the evidence is evaluated as on the evidence itself.