Imagine a restaurant that made the exit door slightly harder to find than the entrance. Not locked. Not hidden. Just tucked behind a column, past the dessert display, with a host who asks if you’re sure you want to leave. Technically you can walk out. Practically, some people order another coffee. That’s the design problem at the center of Amazon’s $2.5 billion settlement with the FTC, and starting October 1, 2026, the compensation for it shows up in accounts automatically, up to $200 per person, with no form to fill out.
I review AI tools for a living, which mostly means I spend my days reading interfaces the way a mechanic listens to an engine. And the thing that keeps pulling me back to this settlement isn’t the dollar figure. It’s that the remedy is the exact opposite of the problem. The alleged offense was friction. The fix is zero friction.
What’s actually happening
The mechanics are refreshingly dull, which is the highest compliment I can pay a consumer refund program:
- Automatic refunds of up to $200 begin October 1, 2026, with no action required from eligible customers.
- Payments are scheduled to be completed by April 2027.
- Amazon began sending payments in November 2025 after the settlement over misleading enrollment and cancellation practices.
- As of September 2026, more than $845 million had already gone out.
- The original settlement covered members who used fewer than 10 Prime benefits.
- If consumer-accepted payments don’t hit the required threshold by February 2027, Amazon has to send additional automatic payments to people who already received refunds.
- A second phase, the claims process, opens after automatic payments wrap. Consumers who used more than three but fewer than 10 benefits in a year may be able to file.
That $845 million number deserves a second look. It means roughly a third of the settlement value moved before the October wave even started. This isn’t a press release about future intentions. Money has been landing in accounts for the better part of a year.
Why an AI reviewer cares about a subscription lawsuit
Because the industry I cover is building the next generation of these flows, and it’s building them faster than anyone can audit them.
Every AI product I test right now has a subscription layer. Most have a free trial that converts silently. A growing number have usage-based pricing on top, where an agent burning through tokens in the background generates charges nobody explicitly approved. The pattern Amazon got fined for — enrollment that’s easier than exit — is now the default architecture of an entire category of software.
And AI makes the asymmetry worse in a specific way. When a support chatbot handles your cancellation request, the company controls the script, the tone, the number of clarifying questions, and how many times it offers you a discount before it finally processes what you asked for. Every one of those is a dial that can be tuned toward retention. The dark pattern stops being a button placement decision and becomes a conversation policy, which is far harder to screenshot and far harder to regulate.
I’ve cancelled AI subscriptions this year that required me to explain, in a text box, why I was leaving before the cancel option appeared. I’ve hit chat agents that looped twice through retention offers. None of that is illegal. All of it is the same family of design thinking that cost Amazon $2.5 billion.
The part that should make founders nervous
Look at the threshold clause. If consumer-accepted payments fall short by February 2027, Amazon owes more automatic payments to people who already got refunded. That structure is worth studying if you build software. It means the company can’t satisfy its obligation by making the money technically available and hoping people ignore it. Low uptake triggers a bigger bill.
Regulators learned something from watching companies design unclaimed settlements. The frictionless-by-default approach, with a penalty for low participation, is a genuinely smart response to a decade of class actions where the claim rate hovered in single digits.
The tiered eligibility is interesting too. Fewer than 10 benefits used gets you an automatic payment. Between three and 10 in a year may get you into the claims phase. Someone sat down and built a usage-based definition of harm. Anyone shipping an AI product with usage-based billing should read that twice, because it establishes that “did the customer actually get value from what they paid for” is a question a regulator is willing to quantify.
What to do with this
If you’re a Prime member, nothing. That’s the point. Watch your account between October and April, and be aggressively suspicious of anyone emailing you a link to “claim” your settlement money. Automatic means automatic. Any message asking for your details is someone else’s business model.
If you build software, treat this as pricing. Two and a half billion dollars is what a cancellation flow costs when you optimize it for the wrong number. The AI tools I’ll be reviewing next year will be judged partly on how fast I can leave them. That metric just got a market rate.
🕒 Published: