A Diabetes Drug Just Quieted Cravings Science Couldn't Explain. Washington Just Deleted the Number That Told You How Much Was Too Much.

Naomi drank vodka around the clock — the kind of drinking that gets described, clinically, as ICU-level, the kind where stopping cold is its own medical emergency. She is not drinking that way anymore. She is not fully abstinent either. She drinks minimally now, and she has her children back. Her story, and nine others like it, ran in Filter Magazine on July 1, collected by Kenneth Anderson, who founded the harm-reduction support network HAMS in 2007 on the premise that recovery doesn’t have exactly one shape.

The country spent this month proving both halves of that premise at once — that craving has a biological answer, and that it still won’t put a number on how much is safe.

If you have ever stood in a kitchen at 9 p.m. arguing with yourself about a drink you already knew you were going to have, you already understand the thing this article is about better than most of the people who are about to be quoted in it. The argument isn’t really about the drink. It’s about a switch flipping somewhere behind your eyes that makes the drink feel less like a choice and more like a debt coming due. Addiction medicine has a name for that switch. Science just got a much better look at how it’s wired. And the same year that happened, the government that funds the science quietly stopped telling you what “too much” means.

The switch isn’t a metaphor — your brain is doing math, and craving changes the math

For most of the last century, a craving got treated as a character problem: weak will, bad choices, a moral failure to be confessed and prayed through. Theresa, one of the ten people in Anderson’s piece, spent years inside that framework before she rejected it. She’s three years sober now, on naltrexone, and she doesn’t use the word “powerless.” “I’m not powerless. I’m in control,” she says — a direct answer to the first of Alcoholics Anonymous’s Twelve Steps, which asks members to admit they are powerless over alcohol. Mary, in the same piece, points to something more specific: she learned about “kindling,” the well-documented phenomenon in which each withdrawal episode leaves the nervous system more sensitized than the last, so the tenth time you try to quit cold can be more dangerous than the first. That single fact — that your body keeps score — was the thing that finally moved her.

Here’s the plain-language version of what a team at Yale just confirmed about that scorekeeping. Researchers led by Xiaosi Gu, director of Yale’s Computational Psychiatry Unit, published a study in Nature Mental Health in which 132 people with moderate-to-heavy alcohol or cannabis use played a slot-machine-style game while reporting their craving level moment to moment. The question wasn’t whether craving made people want the drug more — obviously it does. The question was whether craving changes how the brain learns. It does, and not in the direction you’d expect: in the alcohol group, the stronger someone’s craving, the faster their brain locked onto reward-predicting patterns, almost as if the craving were sharpening the exact learning system that keeps the habit running. In cannabis users, craving did the opposite — it slowed learning down. Same feeling, opposite neural signature, depending on the drug.

“This could explain why breaking the addictive cycle feels so difficult,” Gu said, “as the brain is adapting constantly.” Translate that out of the lab: a craving is not you failing to argue your way out of a bad decision. It is your brain, in that exact moment, getting better at the thing you’re trying to stop doing. You are not losing an argument with yourself. You’re losing a race against a system that was built by evolution to win races like that.

We have been here before, misreading the mechanism as the moral failure. In the 1990s, a woman named Audrey Kishline built an entire movement — Moderation Management — on the idea that not everyone with a drinking problem needed the total-abstinence, one-true-path model that had defined American treatment since the 1930s. She was right that the field needed more than one door. She was also, later, catastrophically wrong about which door was hers: she left the program she founded, drank heavily, and in 2000 killed two people driving drunk. Her story has been used both to indict harm reduction and to make the more uncomfortable case her critics didn’t want to hear — that pretending one model fits everyone kills people either way, just differently. The Yale data doesn’t resolve that fight. It does explain, mechanically, why it’s still a fight forty years later: two different drugs can hijack learning in two different directions, and a treatment system built for one crude average was never going to fit either one precisely.

She was also, later, catastrophically wrong about which door was hers: she left the program she founded, drank heavily, and in 2000 killed two people driving drunk.

A drug built for diabetes just moved the needle nobody could move on purpose

Here is the part that would have sounded like science fiction five years ago: a study published July 21 in BMJ Open found that people with alcohol use disorder who were prescribed GLP-1 drugs like semaglutide — for diabetes or obesity, not for drinking — had a 22 to 26 percent lower risk of alcohol-related hospitalization than people on other diabetes medications, and a 32 percent lower risk than people on other weight-loss drugs. The researchers, led by Hemalkumar Mehta at Johns Hopkins and Jay Lusk at Duke, used de-identified health records from 40,703 adults and deliberately designed the analysis — an approach called target trial emulation — to imitate a randomized trial using real-world data, specifically to rule out the obvious objection that people who get prescribed these drugs are just healthier or more careful to begin with. The effect showed up only for alcohol-related hospitalizations, not hospitalizations generally, which is the strongest evidence they have that this isn’t a fluke of who gets prescribed what.

It isn’t alone. A randomized, placebo-controlled trial out of Copenhagen, published in The Lancet in April put 108 people with AUD on weekly semaglutide plus talk therapy versus placebo plus the same therapy, for 26 weeks. The drug group cut heavy drinking days by 41.1 percent — a 13.7-point advantage over placebo, confirmed against blood-alcohol biomarkers so it wasn’t just self-report. George Koob, director of the National Institute on Alcohol Abuse and Alcoholism, called it plainly: “Very few medications are currently approved for alcohol use disorder… A new option that is more accessible and more effective could be a gamechanger.” Nora Volkow, who runs the National Institute on Drug Abuse, put it in the frame that matters for anyone tracking where addiction medicine is actually headed: “We’re beginning to see some of that potential for GLP-1s to treat drug addiction turn into reality.”

Name the limit here, because a source that hides its own caveats isn’t one you should trust more for it: the BMJ Open study is observational, not a randomized trial, and even its authors are careful to call the finding an association, not proof of a mechanism. Nobody yet knows exactly why a drug that slows gastric emptying and blunts appetite would also blunt the pull of a drink — leading theories point to the same brain reward circuitry both hunger and craving run through, but that’s a hypothesis, not a settled answer. Only four medications are FDA-approved for AUD today — naltrexone, acamprosate, disulfiram, and off-label topiramate — and none of them work for most people who try them. Right now, a drug built to lower blood sugar is outperforming that entire toolbox in early data, by accident, before a single pharmaceutical company designed it to.

The same year the science got sharper, the warning label got vaguer

Here’s the part that should make you angry, or at least suspicious, and it isn’t a coincidence of timing so much as a collision. Since 1990, every edition of the federal Dietary Guidelines for Americans put a number on the page: no more than two drinks a day for men, one for women. That number came out of the same government now funding the GLP-1 research above. The 2025–2030 edition, published in January 2026, deleted it. What’s there now is softer language — “consume less,” “limit” — with no threshold and, critically, with most of the specific cancer and cardiovascular risk warnings stripped out too, leaving only a nod to breast cancer where the prior science covered seven cancer types.

The American Association for the Study of Liver Diseases didn’t let that pass quietly. Their statement cites the 2025 Surgeon General’s Advisory directly: alcohol contributes to roughly 100,000 cancer cases and 20,000 cancer deaths every year in the U.S. AASLD also flagged something the new guidelines don’t mention at all — that men and women metabolize alcohol differently, which is exactly why the old guidance had two separate numbers instead of one vague sentence. The Lancet Gastroenterology & Hepatology called the change what it looked like: a step backwards, arriving in the same political moment as heavy lobbying from distillers, brewers, and vintners against exactly the numeric limits that just disappeared.

Their statement cites the 2025 Surgeon General’s Advisory directly: alcohol contributes to roughly 100,000 cancer cases and 20,000 cancer deaths every year in the U.S.

Thirty-six years of a government telling you, in writing, where the line was. Gone in one edition, the same season researchers finally started explaining why so many people need help finding a line at all. You don’t need a committee’s number to know your own body’s answer. But taking the number away doesn’t make the biology optional — it just means the 27.9 million Americans with alcohol use disorder, only about 7.6 percent of whom get any treatment in a given year, now have one less piece of paper backing up what their own experience has been telling them.

What you already know, and what’s actually new

If you’re the one standing in the kitchen at 9 p.m., none of the neuroscience is going to feel like news. You already know the argument isn’t really an argument — you know how it ends before it starts, most nights. What’s new isn’t that craving is powerful. It’s that researchers can finally point to the exact mechanism doing the overpowering, and that a drug nobody built for you might be quietly doing something about it while the agencies that are supposed to warn you go softer, not harder.

Grace, also in Anderson’s piece, tapered down from a nightly 1.75-liter bottle to drinking socially, maybe once a month, using a structured method inside HAMS. She didn’t wait for the government to hand her a number, and she isn’t waiting for a GLP-1 prescription either. What she had was a framework that didn’t require her to fail at total abstinence before anyone would take her drinking seriously — and time. Naloxone for opioid overdose is still federally funded and still yours to carry regardless of what happens to any dietary guideline; the same is true of naltrexone and acamprosate for alcohol, both still FDA-approved, still prescribable this week by any doctor willing to write it, no new legislation required. The tools didn’t disappear. The warning label did. The distance between those two facts is where the next year of this story gets written — and whether you’re the one drinking, the one who loves someone who is, or the one prescribing the medication, that distance is worth watching closely, not politely.