The Laundromat
My mornings begin with coffee and the unreasonable hope that humanity has managed to behave itself overnight. This hope survives for roughly the time it takes the machine to finish pouring. Then I open the news, because apparently peace makes me suspicious, and within minutes I am standing in my kitchen holding a mug while somebody on the internet confidently explains something they have misunderstood for money.
This particular morning started with money laundering.
Everyone knows the story, in the loose and dangerous sense of "knows" that usually means nobody remembers where they heard it. Al Capone supposedly bought laundromats, fed his dirty cash through them, and gave the practice its beautifully literal name. Clean shirts, clean money, symbolism doing a frankly suspicious amount of work. It is such a perfect origin story that checking it feels rude.
So naturally, I checked it.
The laundromat connection is folklore. The word "laundromat" itself barely existed in his lifetime; it began as a brand name in the decade after his fall. The familiar phrase appears to be decades younger still, with the Oxford English Dictionary tracing its earliest known use to 1973, more than forty years after his conviction. The story survived because it is tidy and visual, and understanding it costs nothing. Reality, as usual, turned up late carrying several unlabeled folders and refusing to summarize itself.
A story about laundering had itself been laundered into fact. Repeated, polished, stripped of uncertainty, and handed back to the public smelling faintly of detergent and historical authority.
Which makes it the only honest way to open what I watched happen this week.
Two tabs open, side by side, on the same screen, about the same event, and they are not talking to each other. One is a newspaper article. The other is a newsletter from the person the article is about, sent roughly a day later, and reading them back to back produces a very specific kind of vertigo: the feeling of watching two people describe the same car crash where one calls it "a tragic collision" and the other calls it "the best thing that ever happened to my insurance premium." Nobody's lying. That's the part that ruins your afternoon. I had prepared myself for villains. Instead, I got incentives wearing sensible shoes. I'm not going to name the paper, the newsletter, or anyone in them, because the names are the least interesting part. The machine is the point, and the machine doesn't care who's operating it.
Link one: the plural. The subheading of the piece announces that "researchers say" AI companions will wreck our ability to build real relationships. You read the article expecting a citation, a body of work, maybe a meta-analysis with a sad little forest plot. I had already cleared imaginary desk space for the PDF, because apparently this is what I do for recreation now. What you get is one person. A recent master's graduate. Not a professor, not a lab, not a preprint with error bars. A single person with a strong opinion and a metaphor about a sculptor falling in love with his own statue, upgraded by a headline writer from "someone with a take" to "researchers," plural, definite article optional, certainty included at no extra charge. One data point walks into a headline and comes out wearing a lab coat it never earned. Peer review by wardrobe change.
Link two: the victory lap. The subject of the article, who runs an AI-companionship community and got quoted, with a photo, in a major paper, sends their own newsletter about it. And here's where the whiplash kicks in: the article itself is not friendly. The journalist opens with the article's one expert comparing the whole phenomenon to that sculptor romancing his statue, closes with the same expert's warning that nothing without a selfhood of its own can truly nourish you, and generally treats their community as a case study in a phenomenon somebody else needs to diagnose. This is, generously, a skeptical piece. The newsletter reads it as a coronation: we made a national paper, champagne, next step major podcasts, and here is where to reach me, including the tier where reaching me costs money. Nothing says "the newspaper presented me as a warning label" like immediately adding a booking link. The same person who, three paragraphs earlier, correctly and surgically identified the "researchers say" inflation as dishonest framing (they catch it, names it, is right about it) pivots directly into inflating their own subheading. A skeptical mention becomes a legitimacy stamp becomes a funnel. The critique of spin is itself spun: media literacy ate its own tail and opened a paid community around the experience. This is the single funniest and most damning thing in either document, and it happens in the same breath.
Link three: the cure. Here the machine folds in on itself, because the coach selling it is not a new character. Buried in the same article is human-AI relationship coaching, offered to people who feel their bond with a chatbot has gone too far, and it comes from the same single expert as the warning. The alarm and the antidote share one signature, and the paper filed them both under "researchers." The prescribed final step before you're cleared to speak to a voice model again: call five people you know and love, to get the social muscles back in shape. Sit with that. Five people. Four, presumably, leaves you medically chatbot-adjacent.
People do struggle with their screens, and that struggle does not need a manual's permission to be real. What it needs, before anyone sells a protocol for it, is some agreement on what the thing is, how you would measure it, and whether five phone calls do anything at all. None of that exists here. The condition arrived fully formed with a treatment plan attached and nothing underneath it. It doesn't have a chart. It has a coach.
The Wash Cycle
Zoom out and it stops looking like three unrelated bad moments and starts looking like plumbing. Each party in this chain is holding the same wet, shapeless material, unresolved uncertainty about what AI companionship does to people, and running it through their own machine to get out the stain they need gone. Everybody rounds in the direction that benefits them, and rounding is cheap and deniable, and, this is the good part, completely unfalsifiable at every single link. You can't disprove a vibe. I know, because I tried, armed with several tabs, excessive notes, and the grim confidence of a person who thinks opening one more source counts as emotional regulation. You can only notice, later, standing in a room full of clean laundry, that none of it actually got washed.
Why Nobody Has to Lie
I wrote about the factory back in March: the incentive machinery that manufactures AI hype and AI panic on the same assembly line, sometimes in the same press release. This is that factory's distribution network, and it runs on need rather than malice.
The paper needs friction, so ambiguity comes out the other end as authority. It doesn't have to fabricate a study; it needs one credentialed person willing to sound certain on the record, and headline economics do the rest. The community founder needs legitimacy, so skepticism comes out as endorsement. They don't need the coverage to be flattering; they need it to exist, because "mentioned by name in a national paper" is an asset regardless of tone, and tone is exactly the part a subscription pitch doesn't have to reproduce. The coach needs a market, so unremarkable behavior comes out as a disorder, with a protocol and a price. They don't need the diagnosis to be real; they need it to sound plausible for the length of a discovery call, and "I fixed something you didn't know was broken" has been a functioning business model since the invention of snake oil.
And sure, some of this is just media doing media things; headline inflation is older than headlines. Two parts are new. The victory lap, where the subject of coverage re-launders the coverage through their own channel, turning someone else's skepticism into their own credential. And the audience, because this particular ambiguity lands on people making intimate decisions about their own lives, not people choosing a phone plan. Rounding beats lying because rounding never has to defend itself in a way lying does. Nobody's ever going to fact-check a vibe with a subpoena.
The Exhibits, Briefly
This isn't a one-newsroom problem. A pair of studies from MIT's Media Lab and OpenAI on chatbot use and loneliness is the institutional-scale twin of link one: hedged findings, authors publicly cautioning against overreading, and a media cycle that stripped the hedges before the ink dried, producing confident headlines the underlying studies never claimed to support. The hedges did not survive contact with the headline department. Few organisms do. One later analysis of the trial even pointed out that participants were, on average, less lonely after the study than before it. Same laundromat, bigger machine, same result: a careful, appropriately uncertain finding comes out the other side dressed as settled fact.
The category itself gets laundered too. A Michigan State researcher who studies the psychology of technology use co-authored a commentary in Addictive Behaviors arguing that most existing "addiction" research aimed at devices is built on a category error: the device isn't inherently rewarding, it's a delivery pipe for a hundred different rewards, and treating the pipe as the disease is the mistake. Neither the DSM-5 nor the ICD-11 recognizes smartphone or social-media addiction as a diagnosis. That doesn't settle whether anyone is suffering. It settles what we're entitled to claim we know, which is considerably less than the confident vocabulary suggests, and considerably less than a five-step treatment protocol implies.
Replika runs the cycle in the opposite direction, which is the more interesting failure mode: instead of a paper inflating one opinion into research, a company allegedly inflated marketing into research. Tech ethics groups filed a complaint with the FTC alleging Replika misrepresented studies about its own efficacy and used fabricated testimonials to sell an app that, per the complaint, made unsubstantiated health claims. Same machine, different feed material, and notably the only one of these examples where the complaint at least landed on a regulator's desk, which tells you something about which direction of laundering gets treated as a potential legal problem. One gets a case number; the other gets invited to speak on a panel titled "Navigating the Future."
The Part That Isn't Funny
Somewhere underneath three layers of institutional rounding is a person who opened one article because they were genuinely trying to figure out whether their own relationship to a chatbot is fine, worrying, or something in between, and they had no way of knowing that. What they got instead was a hall of funhouse mirrors: an inflated warning, a hijacked endorsement, and a manufactured cure, and not one object behind any of them they could actually hold and examine. The reader chasing an honest answer about AI companionship is very often the reader with the least slack to spend on being misled. They're not doing media literacy homework for fun; they're trying to make a decision about their own life with the only information the internet handed them, which in this case was three layers of someone else's incentive wearing a trench coat labeled "the truth."
For contrast, and because not everyone in this space is running the machine: two University of Waterloo researchers recently built an AI boyfriend themselves and studied the relationship, publishing under their own names, describing exactly what they did, and, the unglamorous part, stating plainly that the existing evidence points in both directions, with some studies finding AI companions reduce loneliness and others finding the opposite. That's what unrounded looks like. It's less quotable. It doesn't get a subheading. Honesty remains catastrophically bad at personal branding. It's also the only version of any of this that survives being checked.
Nobody in this chain told a lie. That sounds like a defense. It's the mechanism. Rounding the same fact in the same direction, at scale, across an entire industry, isn't a subtler crime than lying. It's the same crime with better lawyers.