National drinking guidelines now disagree by a wide margin. The evidence is the same.

Ask two English-speaking governments how much alcohol is safe and you will get answers that differ by nearly an order of magnitude. Canada's 2023 guidance says two standard drinks a week or fewer carries low risk [s3]. United States dietary guidance still describes moderate drinking as up to one drink a day for women and two for men [s4]. Both bodies were reading the same research. The disagreement is not about the evidence. It is about what to do with a risk that has no clean cut-off.

The curve has no step in it

The single most consequential finding of the past decade came from the Global Burden of Disease Study, which pooled 694 data sources on consumption and 592 studies on risk to model alcohol's net effect on health [s1]. Its conclusion was blunt: across all health outcomes combined, the level of consumption that minimised harm was zero standard drinks per week [s1]. A standard drink there was defined as 10 g of pure alcohol [s1].

That finding is what "no safe level" refers to. It does not mean one drink is dangerous in the way a cigarette is. It means the risk curve rises from the origin without a protected zone at the bottom — there is no amount that is measurably better for overall health than none [s1]. The study also found that alcohol was the seventh leading risk factor for death worldwide in 2016, and the leading risk factor for people aged 15 to 49 [s1].

The honest caveat is in the numbers themselves. The estimate of the harm-minimising level was zero, but with an uncertainty interval running from 0.0 to 0.8 standard drinks per week [s1]. The data cannot distinguish "exactly zero" from "a very small amount," which is a different and more modest claim than the slogan suggests.

Why the same curve produces different guidance

A continuous risk curve forces a policy choice that the science cannot make: how much risk counts as acceptable, and how to express it. Canada's guidance resolved this by describing a continuum. It places one to two drinks a week at low risk, three to six drinks a week at moderate risk, and seven or more at a level where the risk of heart disease, stroke and several cancers rises appreciably [s3]. At three to six drinks a week, Canada's report states, the risk of cancers including breast and colon cancer begins to climb [s3].

US guidance took a different frame, setting daily limits rather than a weekly continuum, and retaining the older "moderate drinking" category [s4]. The 2025 Surgeon General's advisory explicitly asked that these limits be reassessed in light of the cancer evidence, which is a government office acknowledging that its own country's numbers may be behind the research [s4].

The relative-versus-absolute problem underneath it all

The reason guidance can diverge so much without either side misreading the science is that "risk rises with no safe level" and "the added risk from one drink is small" are both true at once. The Lancet Public Health statement makes the point directly for cancer: risk starts at low levels of consumption, and alcoholic beverages are a Group 1 carcinogen, but the increase in absolute risk at light drinking is modest even as it produces a large number of cases across a whole population [s2].

A guideline writer must decide whether to emphasise the slope, which never flattens, or the small size of the individual step, which stays small at the bottom. Canada emphasised the slope; the United States, historically, emphasised the step. Neither is factually wrong. They are answering slightly different questions.

What this means for reading any number you are given

The practical upshot is that a drinking guideline is not a measurement, the way a speed limit is not a claim about where cars crash. It is a judgement about acceptable risk, dressed in a number. When guidance changes — as Canada's did, dropping from a previous limit of around ten drinks a week to two — the underlying biology has usually not changed [s3]. What changed is the tolerance for risk built into the recommendation, and the weight given to cancer relative to the older, and now largely discredited, idea of a cardiovascular benefit.

The most defensible reading of the current evidence is the least satisfying one: less is lower risk at every point on the curve, the curve probably passes through or very near zero, and the exact shape at the very bottom is genuinely uncertain [s1][s2]. Anyone who tells you the science has settled on a precise safe threshold is describing a policy choice, not a finding.

This article describes what the evidence and the guidelines say. It is not medical advice.

Sources

• Alcohol use and burden for 195 countries and territories, 1990-2016: Global Burden of Disease Study 2016 — The Lancet, 2018-08-23
• Health and cancer risks associated with low levels of alcohol consumption — The Lancet Public Health, 2023-01-01
• Canada's Guidance on Alcohol and Health: Final Report — Canadian Centre on Substance Use and Addiction, 2023-01-17
• Alcohol and Cancer Risk: The U.S. Surgeon General's Advisory — US Surgeon General, HHS, 2025-01-03