Nobody Asked

There’s a company in New York that will clean your apartment for free. The cleaner arrives wearing a small camera on their head, and while they wash your dishes and fold your laundry, it records everything. That footage trains household robots to do the work someday. The cleaning is free because the recording is worth more than the labour.

It’s a fair offer, on its face: they blur faces, they tell you what they collect. You get a spotless apartment; they get training data.

I notice because I spent years on both sides of this trade: CTO of 1X, one of the companies now putting a humanoid robot into homes, then CTO of a data-labelling company built to feed AI’s appetite for data. I know that hunger can’t be sated.

So here’s a test. Any deal for your data owes you three things. Benefit: you get something real back. Transparency: you’re told plainly what’s taken. Agency: you understand it well enough to say no. Notice which leg does the work. Benefit can be real and the deal still rotten; transparency complete and still buried. Agency decides whether it was a deal at all: a benefit you never got to refuse isn’t a trade, it’s extraction. Run the free cleaning through it: a camera in your home doesn’t only capture chores, it captures the symbols of your faith and the prescription bottles in your cabinet. You bought the cleaning; the rest was the product too, and no one showed you that part, so you couldn’t refuse it: not its scope, not for everyone else your camera caught who never signed anything. And that’s the honest version; most of what’s happening doesn’t get even that far.

But the friction is gone. General Motors sold hundreds of thousands of drivers’ data to brokers who score them for insurers; some saw premiums jump or coverage denied, and learned a file existed only when reporters exposed it. This year the bill came due: a federal ban and the largest penalty ever under California’s consumer-privacy law. In May the FTC settled with a media group whose AI ad service, supposedly built on what smart devices overheard at home, turned out to rest on resold broker lists; the firms never listened at all, and the order rejected clicking through terms of service as consent to collect from inside a home. Different industries, the same missing piece: nobody was in a position to say no. Abundance without choice isn’t progress; it’s extraction with better marketing.

Automation was supposed to lift us up, to take the work that’s dull, dangerous and dirty and free us for something better. We’ve quietly run it in reverse: when your phone, your car, and your kid’s tablet are all sensors, the job left to you is to demonstrate your own life so a machine can learn to live it. Every automation wave looked like a raw deal from inside, and the doomsayers lost; what’s different now is that you can’t see the trade, so you can’t refuse it. I know, because I helped build it. The home robot 1X promised to ship this year took, the company says, more than ten thousand pre-orders at up to $20,000 each; as of late July, no customer has one. For the hard parts, a remote worker in a headset steers it while its cameras record your home. “If we don’t have your data, we can’t make the product better,” the CEO told the Wall Street Journal. The trade is disclosed, and buyers still choose it, so this isn’t a stolen yes. It’s scope: you agreed to be filmed folding laundry; your houseguests, your kids, the prescription on the counter did not.

A lot of people don’t care. “I have nothing to hide,” they say, and they aren’t foolish. They’re tired. So take the secret ballot: we curtain it not because everyone has something to conceal, but to protect the one person who’d be fired or harassed for their vote, and that holds only because the booth is private for everyone. Privacy shelters anyone only when it covers everyone. And they are real. In Nebraska, police got a warrant for a mother and daughter’s Facebook messages; Meta handed them over, and the chat where the two discussed ending a pregnancy became key evidence. You may have nothing to hide. But the next woman deciding whether to type that question into a chatbot does, and her safety depends on privacy being ordinary. Give a little away because it felt harmless, and you thin the cover she stands under.

You can grant all that and still think privacy loses on economics: scale wins, and restraint looks like a handicap. But if the only thing that pays is collecting everything, restraint never arrives on its own; it has to be built into the architecture, not bolted on as a promise. And the handicap is narrower than it looks: distrust already prices that cost, and in the parts of life that touch you, a system that runs on your device is often the better product. The difference that matters is between two kinds of architecture. The custodial kind collects the data and tries to contain it; the subtractive kind never collects the sensitive thing, so there’s nothing to contain. Watch the first fail. WIRED found a complete face-recognition pipeline sitting dormant in Meta’s smart-glasses app (built, switched off, shipped to fifty million phones), waiting on nothing sturdier than a promise Meta can revoke. Nobody bought a phone to host that, and nobody was asked. The safeguard is only that they said they wouldn’t turn it on. Build it the other way: the sensitive thing never collected, and there’s nothing to leak, subpoena, or sell.

Agency was the leg that went missing, and the instinct is to want it back: the click, the consent. That instinct is right but doesn’t go far enough: a yes you can give is a yes you can be talked out of, or that quietly stops being offered. The refusal that survives is built in before anyone gets to ask. Nothing makes betrayal impossible: there’s always metadata, the model that infers what you didn’t say; “made impossible” is really made expensive, made pointless, removing the thing there is to betray. And not every trade is a trap: the free cleaning, taken with eyes open, is a real choice a good system leaves you. The enemy was never the trade. It’s the vanished no.

The choice is still up for grabs, surrendered one refusal at a time, and it isn’t yours to make alone, which is why the rules have to carry the weight a click never could: collect less by law, keep the off switch by law, make whoever holds your data carry the cost when it leaks. So when someone wants your trust along with your data, whether they’re selling a robot for your kitchen or a camera they promise to keep switched off, don’t ask what they’ve promised. Ask the only question that holds, and make the law ask it too: what have you made impossible?