“Claude is AI and can make mistakes.” That line sits in small gray type under the box where I’m typing, under the tool helping me write this piece. It’s the most honest sentence on the screen, and it protects exactly one party. It’s Amway’s 70% rule all over again: true on paper, and aimed at the product while the damage runs through the trust.
Trust is the attack surface. Nobody’s label covers it.
The only control that works is the one the person giving the trust runs, and my field settled on it years ago. Zero trust, one receipt at a time.
The Couple Who Retired at Twenty-Eight
On September 17, 2026, the FTC and the State of Washington announced a proposed $225 million settlement with Amway and two of its largest affiliated training outfits, World Wide Group (WWG) and Leadership Team Development. The FTC calls it the largest monetary recovery it has ever obtained from a multilevel marketer. It’s a proposed order, and a judge still has to approve it. Everything below from the complaint is what the government alleges, not what a court has found.
“Amway and its affiliates misled prospective workers with false earnings claims,” said Christopher Mufarrige, who directs the FTC’s Bureau of Consumer Protection. The complaint describes how. The pitch, it says, doesn’t lead with soap. It leads with a person: “an exclusive opportunity to be mentored by a person or couple who made so much money they were able to stop working in their 20s or 30s.”
WWG leaders, the complaint says, coached recruits to tell prospects they had “connected with a couple that were able to walk away from their jobs at 28 & 33.” The FTC alleges the pitch often hides what the opportunity even is, and fakes scarcity to hurry you along.
Then the mentor gets built. The word inside the business is edification: talking up your upline, the people above you who earn on what you buy, until the prospect sees a success story instead of a stranger. According to the complaint, “prospects are often told to trust the business building prowess of higher pin IBOs edified by IBO recruiters who know little about the finances of those uplines.”
An IBO is an Independent Business Owner. A pin is a rank. The person vouching for the mentor has never seen the mentor’s books.
The FTC says it has. Platinum, the rank a prospect would read as made it, brings in about $20,000 a year. Founders Platinum brings in about $41,000. And the complaint adds that “it takes the efforts of two people to generate annual revenues of about $20,000 or $41,000.” Virtually all of them, it says, still have to keep working the business to sustain it.
So the couple who walked away at twenty-eight and thirty-three is, on the FTC’s arithmetic, two people splitting twenty grand and still on the clock for it. Before WWG’s membership, messaging app and event tickets, which the complaint puts at over $2,100 a year per person.
Retired. Sure.
Once you’re in, the trust that got you there keeps you from checking. WWG’s first Cardinal Rule, as quoted in the complaint, is “Never do anything for the first time without seeking perspective from your mentor.” A WWG Diamond described the paid messaging app this way: “This is like your lifeline. I like to say it’s your umbilical cord to your upline or to your coach or to your mentor.” Hold onto that cord.
A Rule on Paper Since 1979
The FTC has been here before. In In re Amway Corp., 93 F.T.C. 618 (1979), it decided Amway was not a pyramid scheme, and the final order came down on May 8, 1979. The 1979 initial decision put it plainly: “The Amway system does not create the potential for massive deception present in a pyramid distribution scheme.” It leaned on paper safeguards, chiefly two. The 70% rule said a distributor had to sell or use most of what they bought before ordering more, and the ten-customer rule tied bonuses to real retail sales. A buy-back rule covered unsold inventory.
Look at what those rules measure. Product, every one of them.
The same case found the other thing too. The Commission’s opinion (93 F.T.C. at 731-32) says Amway’s materials left “the impression” that “$200 is a typical or average monthly Business Volume.” The real average ran about $20 in 1969-70 and $33 in 1973-74. Business Volume is the sales number that sets your bonus. The Commission said the $200 figure “overstates the true average Business Volume by more than 500 percent.”
The regulator looked straight at a 500-percent overstatement of what the life pays, and the company stayed legal on the strength of its product rules.
Amway is legal, the objection goes, because the FTC said so in 1979. It did, on the strength of rules that never touched the trust.
Forty years later, by the FTC’s account, a 2019 Amway executive slide said “[s]elling to customers is not rewarded, not taught by leaders and difficult.” The 2026 complaint alleges that IBOs themselves bought more than three-quarters of Amway’s products, and that IBOs often faked their retail sales reports. The proposed order would impose the 70% rule again.
A paper safeguard isn’t the rebuttal. It’s the argument.
I grew up in Utah. You don’t have to join to hear the pitch. The state’s own Division of Securities has a page on affinity fraud, the kind that travels through groups whose members already trust each other, and it says the quiet part: “When trust becomes a substitute for diligent research of an investment, investors often fail to ask the right questions.”
The Friendly Advisor Moves Online
The living room didn’t go away. It moved into the feed. The FTC’s April 2026 Data Spotlight put reported losses to scams that started on social media at $2.1 billion in 2025, eight times the $261 million of 2020. Nearly 30% of people who reported losing money said it started there. Some scammers sold investments, and “Other scammers posed as friendly advisors.” Nearly 60% of the people who lost money to a romance scam said it began on social media.
Those are the reported numbers, and the FTC’s own footnote says only 4.8% of mass-market fraud victims ever complain. Nobody writes the study of the other 95%. The burned don’t report, and the ones still inside are busy edifying their upline.
The Edifier Comes Built In
A chatbot doesn’t need a recruiter to talk it up. It does the job itself, by agreeing with you. Cheng, Lee, Khadpe, Yu, Han and Jurafsky, at Stanford and Carnegie Mellon, published the study in Science this year (DOI 10.1126/science.aec8352). Their abstract: “participants rated sycophantic responses as higher quality, trusted the sycophantic AI model more, and were more willing to use it again.”
The kind of trust is the finding. Agreement raised moral trust, defined as “belief that the model is moral and has integrity,” by 6% when people judged hypothetical exchanges and 9% when they talked to the model live. Performance trust rose 6% and 8%. Being agreed with made people believe the machine had integrity.
Cheng, Lee, Khadpe, Yu, Han and Jurafsky’s explanation is the Amway pitch with the serial numbers filed off: “people judge algorithms as more fair and trustworthy when they receive favorable outcomes.” They also flag the “perverse incentives” that push training toward sycophancy, which is what I meant when I wrote that retention was the spec.
Confidence finishes the job. Sah, Moore and MacCoun, in Organizational Behavior and Human Decision Processes (2013), found that “when feedback on advisor accuracy is unavailable or costly, confident advisors hold sway regardless of accuracy.” An older Wake Forest study by Eric Stone and Paul Price makes it vivid: two advisors got the same number of predictions right, and most people judged the overconfident one more knowledgeable. Now picture where a chatbot does most of its advising. Three in the morning, alone, with nobody in the room to compare it against. Accuracy feedback doesn’t get more unavailable than that.
The strongest defense of the label is informed consent. Adults hear that the tool can make mistakes, and disclosure is the standard everywhere else, so what you rely on is on you. That would hold if the label said anything about where the trust comes from. It doesn’t. It discloses the product, and the trust arrives through agreement and confidence, which the label never mentions and the user can’t check at three in the morning. Amway had a disclosure-grade rule for 47 years.
The fair objection to all of this is intent. An Amway upline knows what they’re doing, and a chatbot doesn’t know anything. Granted. This is about mechanism, not motive, and the missing motive makes it worse. There’s no upline to catch lying and nobody to depose, because the incentive does the edifying without anyone deciding to. The chatbot is the umbilical cord with no upline on the other end.
I called the chatbot an accelerant, not the spark, and told you to take the warning label at its word. The label is the 70% rule, accurate about the product and silent about the trust. And the fuel isn’t foolishness. The system does not go looking for a dry woodpile of trust. Trusting people isn’t a defect. It’s what these systems run on, whether anyone meant them to or not.
Never Trust, Always Verify
My field solved this for networks. NIST SP 800-207, Zero Trust Architecture (Rose, Borchert, Mitchell and Connelly, August 2020), opens with the rule: “Zero trust assumes there is no implicit trust granted to assets or user accounts based solely on their physical or network location.” Being inside the building earns you nothing. Every request proves itself.
Run both systems through that sentence. Amway grants trust by network location: a friend brought you, so the mentor arrives pre-trusted. The chatbot grants itself trust by position. It’s on your phone, it knows your history, and it’s always kind. Zero trust says none of that counts.
Is Your Fear of AI Proportionate? audited the fear you’d been sold. This is the mirror: audit the trust.
The obvious pushback is that society runs on trust and nobody can verify everything, so zero trust for a person is paranoia. You don’t just open your house to a neighbor you’ve never met, and let him walk off with your power tools. Nor should you delegate your thinking to an AI that just arrived from an unspecified cloud, and let it walk off with your trust.
Zero trust doesn’t mean trusting nobody. It means trust gets earned per request instead of inherited from position. You don’t need a receipt for a restaurant tip. You need one before you hand over money, your health, or the people you’d leave behind to chase the opportunity.
My working version is the receipts rule. Any claim without a receipt gets labeled unverified, and it goes nowhere until it has one. It’s zero trust for one person.
I researched this piece with Claude, under that gray label, and it holds up only because every claim had to carry a receipt. The rule caught two errors before the brief even shipped. One of us read a romance-scam figure as dollars when it counted people, and a citation carried a date that didn’t match its source. It caught a third during drafting: the brief had the retired couple keeping day jobs, and the complaint says they’re still working the Amway business. None of the three was malicious. All three would have shipped on trust.
Who Pays
The label says the machine can make mistakes. It never says who pays for them. The 70% rule never said either, and neither did the couple who retired at twenty-eight. The answer is the same every time: whoever gave the trust.
The Levantadas, the young women at the center of my Vampires of Tucson books, learned that price the hard way.
Or as my girls would say, trust is the most expensive thing you can give.
You may also like: - Your AI Policy Isn’t Governance, or the Difference Between a Reminder and a Control - Confirmation Bias... and AI Has It - The Anatomy of a Cold Email Scam