How injecting a peptide you bought online became normal, and the three times America has watched this movie before.
Reid Robison, MD, MBA · September 2026 · Reading time about 25 minutes, or skim the TL;DR and the timeline
Watching trusted people succeed with something lowers how dangerous it feels, even when nothing about the vial has changed. I call this social de-risking, and it is the engine of the gray-market GLP-1 moment.
The drugs are real and remarkably effective. The problem is that the perceived category of the behavior slid from “prescription biologic” to “peptide protocol” faster than anyone could verify what was in the newer vials. Amphetamine diet pills (1940s–60s), Miltown and Valium (1955–80), and fen-phen (1992–97) each ran the same social script.146
The moment we are living in feels unprecedented. From a century up, it is the fourth run of a very specific American pattern: a legitimate weight or mood drug, dramatic and visible results, and a public that decides, mostly by watching each other, what counts as normal use. Start with the big picture, then tap an era to land in it.
Bigger dots are the turning points. The dashed amber line is today. Drag sideways to move through time; pinch or use + and − to change altitude.
Lanes: each colored band is one era. Gray dots are drugs that had their own smaller cycle in between.
Three things jump out from altitude. First, the gaps are shrinking: about thirty years from amphetamine tablets to their scheduling, thirty from Miltown to the Valium hearings, five for fen-phen, and the GLP-1 story has run from approval to gray market in about four. Second, every era ended not because the drug stopped working but because the culture renegotiated the risk it was willing to carry. Third, the drug that never had its reckoning, phentermine, is still the most prescribed anti-obesity medication in the United States, which is a useful reminder that not every cycle ends in a ban.36
Sociologist William Ogburn called it cultural lag: technology arrives faster than the norms that tell us how to use it.21 Here is each era laid over the same six beats. The colored beats happened; the dashed one is where the current chapter is sitting.
Amphetamine entered medicine as a decongestant and a treatment for narcolepsy and low mood, and within a year clinicians noticed patients losing weight.3 By 1962 U.S. manufacturers were producing roughly eight billion 10-mg doses a year, around 43 for every American, and by the late 1960s FDA officials believed as much as half of the supply was leaving medical channels.12 The diet-clinic “rainbow pill” industry, which bundled amphetamine with thyroid, digitalis and diuretics, produced deaths and a 1968 crackdown before the Controlled Substances Act finally reset the norm.35
The closest historical twin to today, in social mechanics if not pharmacology: friends and family saw the results, a quasi-medical supply chain grew to meet them, and the line between treatment and habit blurred for a generation.
Meprobamate was the first blockbuster “minor tranquilizer,” and it moved from psychiatry into ordinary life within about eighteen months of launch.45 Valium was the best-selling drug in America from 1969 to 1982, peaking above two billion tablets a year, before dependence and withdrawal forced the culture, and then the schedule, to catch up.4
The lesson here is about category drift: once a drug is framed as a technology for managing everyday life rather than treating an illness, the threshold for using it drops for everyone.
A small 1992 study series suggested combining two older drugs, and the off-label pairing became a national craze with an estimated 18 million prescriptions in 1996.7 In 1997 a Mayo Clinic series described unusual valvular heart disease in 24 users, and fenfluramine and dexfenfluramine were withdrawn that September.68 Phentermine, the other half, was never implicated and is still widely prescribed.36
This is the fastest polarity flip on record: perceived risk went from “everyone’s doing it” to “get an echocardiogram” in a single news cycle. It shows how brittle social de-risking is when a vivid harm finally appears.
Semaglutide produced about 15% weight loss in STEP 1 and tirzepatide about 21% in SURMOUNT-1, results no prior obesity drug came close to.910 Then demand met a branded price of roughly a thousand dollars a month, insurance friction, and a shortage that legally opened the door to compounding. When the shortage was declared over in 2025, the compounding exemption narrowed, but the habits, telehealth infrastructure and overseas peptide supply did not disappear. By 2026 FDA was warning about vials of semaglutide, tirzepatide, and unapproved next-generation molecules sold online as “for research purposes only.”11
What is different this time: the drug class is far safer and better studied than amphetamine or fenfluramine, and the most likely “reassessment” may be economic (cheaper branded and oral options) rather than a ban. What is the same: the social machinery of de-risking, running at internet speed.
Nothing about the buyer changed. A person who in 2021 would have laughed at the idea of reconstituting a powder from an unmarked vial now does it on a Sunday night while watching television. What changed is the social field around the decision. Try the two dials.
An illustration, not a measurement: felt risk falls with each successful peer you observe, while the product’s risk depends on the product. Bikhchandani and colleagues formalized this as an informational cascade, where later deciders rationally copy earlier ones instead of inspecting the evidence themselves.22
Each label moves the molecule into a different mental neighborhood, and the neighborhood carries its own risk assumptions. “Peptide” lives next to creatine and collagen; nothing about the manufacturing did.26
We infer what is safe from what people like us actually do, far more than from what institutions say we should do. A friend six months in and glowing is vivid; an FDA notice about endotoxins is abstract.23 On TikTok the descriptive norm is set almost entirely by individuals: 95% of the most-liked UK Mounjaro videos came from accounts with no stated credentials, and across thousands of GLP-1 videos only a small fraction detailed any risk.1415
We judge how likely something is by how easily examples come to mind. Thirty pounds lost is unmissable; a subpotent batch, a contaminated vial or a slow biliary problem is invisible at dinner. The mind counts what it can see.25
Most drugs work invisibly. You cannot look across the table and know someone’s LDL is 45. GLP-1s advertise on the body, so every successful user is an unblinded case report with the credibility of a friend. Christakis and Fowler showed body-size norms travel through social networks; whether that reflects contagion or shared environment is debated, but the mechanism they proposed fits this moment uncomfortably well.1617
Many people privately think mixing bacteriostatic water into an unmarked vial is a little insane. They hear friends discuss “units” without hesitation, assume everyone else knows something they do not, and go quiet. The private doubt never gets voiced, so the group looks more confident than any member of it.24
The rigor behind Wegovy and Zepbound (NDA, cGMP, cold chain, decades of trials) casts a halo over anything with the same molecule name. The halo does not check whether the vial was made under USP sterile standards or in a garage. Same word, radically different quality assurance.26
Beyond the visible body change, users describe an internal quiet: the intrusive food thoughts stop. That subjective report spreads as language, and it recruits people who were never trying to lose weight at all.34 In the 1950s, Miltown spread the same way, on the testimony that it made ordinary life feel easier.4
Rogers described how adoption of a new technology moves from a few innovators to early adopters and then, at a tipping point, to the early majority.20 In a friend group, the first adopter pays the whole psychological risk premium: Does it work? Will I get sick? Is injecting myself weird? Once that person succeeds, the questions appear answered experimentally, and the polarity of the decision flips from “why would I?” to “why wouldn’t I?”
Medicine studied this on itself. The classic Coleman, Katz and Menzel study of how physicians adopted tetracycline in the 1950s became the foundation of network diffusion theory.18 A later reanalysis found that much of what looked like doctor-to-doctor contagion was actually aggressive pharmaceutical marketing.19 That caveat transfers directly: in 2026 the friend testimony, the algorithm, the telehealth advertising and the vendor Discord are all pushing at once, and it is not possible from the inside to tell which force is moving you.
The reference point for acceptable pharmaceutical risk is being renegotiated in real time by social networks, not by pharmacology.
Social proof is not a soft, cultural thing sitting on top of a rational brain. The imaging work suggests it is implemented in the same circuits that learn from reward and punishment. Tap each stage.
Puncturing your own skin with an unverified substance normally recruits the threat system: amygdala and anterior insula activity, autonomic arousal, and a visceral “no.” This is not neurosis; it is the default setting that keeps most people from injecting research chemicals. Everything that follows is the story of how that setting gets overridden without a single new fact about the chemical.
The stages are a model, assembled from separate experiments on fear learning and conformity. Nobody has scanned a peptide buyer.
Fear is learned and unlearned socially. Olsson and Phelps showed that watching another person receive a shock paired with a cue produces a conditioned fear response in the observer that looks, physiologically, like first-hand conditioning.30 The reverse also works: Golkar and colleagues found that observing a calm demonstrator with a previously feared cue produced vicarious extinction that was more durable than ordinary extinction, blocking the return of fear.31 A friend injecting at brunch and being fine is, in these terms, an extinction trial for everyone watching.
Disagreeing with the group registers as an error. Klucharev and colleagues found that when a participant’s judgment conflicted with a group’s, activity rose in the posterior medial frontal cortex and dropped in the ventral striatum, the same signature the brain produces after a mistake in reinforcement learning, and the size of that signal predicted how much the person subsequently adjusted toward the group.27 Berns had earlier shown that going along with a wrong group answer changed activity in perceptual areas, while independence carried an amygdala cost.29
Others’ opinions change what things are worth to you. Campbell-Meiklejohn’s group showed that learning “expert” reviewers agreed with you increased ventral striatum activity, and that the strength of that signal predicted how far your own valuation moved toward theirs.28 Apply that to a vial: once the trusted group values it, the internal price of not adopting rises.
GLP-1 receptor agonists reduce intake through receptors in the hypothalamic arcuate nucleus and brainstem, and human imaging shows they dampen responses to food cues in reward-related regions including the insula, amygdala and striatum.3233 That is the pharmacology behind “the food noise stopped.” It is worth sitting with the symmetry: the social field lowers the brain’s risk signal around the vial, and the vial lowers the brain’s reward signal around food. Two valuation systems being tuned at once, only one of them by a molecule anyone measured.
“I’m on a GLP-1” now describes three products that share a molecule name and almost nothing else. The social category collapsed them; the manufacturing did not.
The “units” on an insulin syringe are volume marks calibrated for U-100 insulin. They say nothing about how many milligrams of peptide are dissolved in that volume, which depends entirely on how a given vial was made or mixed. FDA has received reports of hospitalizations after people drew 5 to 20 times the intended dose while converting between milligrams, milliliters and units.12
The paradox in one sentence: objective uncertainty rises with each step away from the regulated supply chain, and subjective risk falls with each friend who takes that step. FDA’s adverse-event counts, 990 for compounded semaglutide and more than 730 for compounded tirzepatide as of May 31, 2026, do not establish causation and are almost certainly incomplete, because most state-licensed compounders are not required to report.11 No comparable count exists for the research-vendor tier, because nobody is counting.
Every previous era closed with the culture catching up to the chemistry. The question is what form catching up takes when the drug is, unlike speed or fenfluramine, mostly good.
A vivid, attributable harm cluster from a gray-market product, a news cycle, and the social polarity flips overnight. The pharmacology of the class makes a class-wide version of this unlikely; a vendor-specific version is entirely plausible.
Slow accumulation of concern, hearings, and tighter rules on compounding and telehealth prescribing. Already partly underway with the end of shortage-era compounding exemptions and the 2026 warning letters.
Price and access collapse the gray market from below: cheaper branded programs, oral formulations, eventual generics. The vials disappear not because people got scared but because a pen at the pharmacy became the path of least resistance.
My honest read as a psychiatrist who has watched a few of these cycles up close: the molecules will be fine, and most of the people will be fine, and the story worth remembering is not about GLP-1s at all. It is that we now have a demonstrated mechanism by which a peer network plus an algorithm can move an entire population’s risk threshold for self-administered pharmacology in about three years. The next thing to ride that mechanism may not deserve the halo.