ALARA and dose limits

Radiation protection at a nuclear facility rests on two requirements that are not interchangeable. The first is the dose limit: an annual figure you may not cross, measurable with a dosimeter and enforceable as a violation. The second is optimisation, which US regulation calls ALARA, for as low as reasonably achievable. ALARA sets no number at all: it requires doses to be driven as far below the limit as is practical, weighing the state of technology, the cost of an improvement against the benefit to public health, and other societal and economic factors. The definition sits in 10 CFR 20.1003 and the obligation to run a program that applies it in 10 CFR 20.1101(b). That is how a plant can be well under the limit and still draw a finding from the regulator: the limit was met, the optimisation was not.

The ceiling itself is public, in 10 CFR 20. For an adult worker the annual limit is a total effective dose equivalent of 5 rem, that is 0.05 sievert, with separate limits of 15 rem to the lens of the eye and 50 rem to the skin and extremities. For a member of the public the limit is 0.1 rem, 1 millisievert per year, plus a cap of 0.002 rem in any one hour in an unrestricted area. Those limits count only what the facility contributes: they exclude natural background, which averages about 2.4 millisievert per year worldwide, and any medical dose the person has received.

Outside the United States the ceiling has a different shape. The IAEA's International Basic Safety Standards, GSR Part 3, set the worker limit at 20 millisievert per year averaged over five consecutive years, that is 100 millisievert in five years, with a cap of 50 millisievert in any single year; for the public, 1 millisievert per year. The US single-year limit is twice the international average, and the international one lets a high year be offset by low ones. The same standard introduces a third quantity that is often mistaken for a limit and is not one: the dose constraint, a value below the limit set in advance for a particular source, used as a tool of optimisation rather than as a legal boundary.

ALARA exists because the model used to estimate the risk has no threshold. The linear no-threshold model assumes the probability of a radiation-induced cancer rises in proportion to dose, with no minimum below which the risk is exactly zero; it extrapolates downward from populations exposed to high doses, where the effect is actually measurable. On that assumption, meeting the limit does not discharge the obligation, because every millisievert avoided still counts, and optimisation follows. The model does not say the dose must reach zero: it says the benefit of reducing dose never vanishes, and it is the word reasonably that applies the brake. The technical argument of recent years is that the model is conservative at the low end, where epidemiological studies lack the statistical power to tell a small effect from none.

On 15 July 2026 the NRC published a proposal that writes that argument into the rule. Docket NRC-2025-1140, at 91 FR 43456, would strike the ALARA requirements from all of 10 CFR Chapter I and replace them with a graded approach to dose management, using determinate thresholds instead of case-by-case judgement, and adds a new mechanism, 10 CFR 20.1205, that would let a licensee authorise a worker above the annual limit provided the worker receives no more than twice that limit in a year and a multi-year average stays bounded, a structure close to the IAEA's five-year averaging. The NRC grounds the change in the linear no-threshold model leading to conservative implementation at low doses, and cites a 2020 analysis concluding that if a threshold for stochastic effects existed it would not be above 1 rem, well under the standing occupational limit of 5 rem per year. Its regulatory analysis estimates industry savings of about $9.53 million per year at a 7% discount rate. Public comment closed on 31 August 2026 and the proposal is not in force.

To read any argument about this, it helps to keep track of what moves and what does not. The 5 rem and 1 millisievert per year limits stay where they are in the proposal: what is under discussion is who decides how far below them you must go, and on what basis. A change to the optimisation is not a change to the ceiling, and the reverse is equally true.