A BBC reporter let an AI company clean his New York City apartment for free. The catch was that two workers wearing camera-mounted hats recorded every surface, every object, and every movement — footage that will train the humanoid robots designed to replace those same workers. The BBC investigation published June 20 is the first first-person account of Shift’s data-for-services model in action.
🔍 THE BOTTOM LINE
Shift isn’t a cleaning company. It’s a data harvesting operation that uses free domestic labour as the acquisition channel. The cleaners are human now. The business model depends on them not being human later. Every free apartment scrubbed is a training dataset sold to robotics companies — and the people opening their homes have no idea what that data is worth, or where it ends up.
What Changed
We first reported on Shift in May when the German startup launched its free cleaning service in NYC. The BBC piece is the next chapter: reporter Archie Mitchell actually let the team into his Upper East Side apartment and documented the experience.
The details are striking. Two mid-twenties college graduates arrived wearing cameras attached to their caps, connected via cable to their phones. They cleaned about five apartments a day, five days a week. They were “intensely focused on their hands” — because the data being captured is specifically about manual dexterity, the precise movements robots need to learn to replicate human physical tasks.
Shift founder Bercan Kilic told the BBC the goal is “to advance humanity.” He pointed to ChatGPT as a model: just as language models learned from existing text, robots need to learn from existing human movement. “In the real world, every object is different, the lighting is different and nothing is the same as it was a couple of hours earlier,” Kilic said. “Models need to learn how their hands, cameras and environments work together.”
The company has expanded beyond NYC. Shift now also has mechanics fixing cars in Turkey — the same model, different skill domain. Free service, recorded data, sold to robotics companies.
The EFF Calls It “Data-Bribing”
The Electronic Frontier Foundation’s Rory Mir, director of open access and tech community engagement, told the BBC the practice is part of a “concerning increase in ‘pay-for-privacy’ and ‘data-bribing’ practices from companies.”
“While it might come with money or a service upfront, the data you share has a way of coming back to bite you,” Mir said. “Even if you trust the business collecting it, there is always a risk of them sharing that information with other businesses or governments.”
Mir added: “We have just lived through decades of our data being used to manipulate us with advertising and predatory practices like surveillance pricing.”
The EFF’s framing is precise. This isn’t a transaction — it’s an extraction. The consumer values a clean apartment at maybe $150. The data collected from that apartment — every surface, every object, the layout of a real home, the physical movements of real cleaning — is worth orders of magnitude more to a robotics company training a $30,000 humanoid. The price is set by the consumer’s ignorance of the data’s value.
NZ Angle
Would Kiwis accept a free cleaner if it meant their home was recorded? The cultural reflex is “she’ll be right” — but the Office of the Privacy Commissioner has been increasingly vocal about data sovereignty concerns. NZ’s Privacy Act 2020 requires informed consent for data collection, but “informed” is doing heavy lifting when most people don’t understand what their home footage is worth to a robotics company.
The question isn’t whether NZ will see similar services. It’s whether the regulatory framework will catch up before they arrive. The Privacy Commissioner flagged AI as a priority in 2025, but no specific guidance has been issued on home-data-for-services models. Shift’s expansion path — NYC to Turkey, with obvious scalability to any urban market — means the question is when, not if.
This is the same pattern we tracked in our reporting on the 5-million-robot workforce: the physical AI revolution needs physical data, and the cheapest way to get it is to offer something for free. It’s also the surveillance concern Meredith Whittaker raised about agentic AI — the gap between what people consent to and what they understand they’re consenting to.
The Other Side
Shift’s argument is straightforward: the exchange is transparent. You’re told upfront that the cleaners wear cameras. You’re told the data trains robots. You can say no. Nobody is forced.
Kilic’s framing — “to advance humanity” — positions the data collection as a collective good. Robotics can’t progress without physical data. Labs and simulated environments only go so far. Real homes, real objects, real lighting, real mess — that’s what makes a robot useful. Without it, domestic robots remain in the demo stage forever.
There’s a version of this argument that’s genuinely democratic: if the data is anonymised, if the trade is explicit, if the consumer has genuine choice, then it’s a voluntary exchange between consenting parties. The EFF’s counter is that “genuine choice” is a fiction when the value asymmetry is this large and the consumer’s understanding of what they’re trading is this low.
The Bigger Picture
The Shift model is a glimpse at the next phase of the AI data economy. The first phase scraped the internet — text, images, code. That’s nearly exhausted, and it’s the subject of ongoing litigation. The second phase is physical: recording human movement, human environments, human dexterity. The data isn’t on the internet. It’s in your home, in your kitchen, in the way you fold a towel.
Companies will offer free services to get it. Cleaning, cooking, maybe eventually childcare or eldercare — any domestic task that a robot might one day perform. The consumers who accept will be, overwhelmingly, the ones who can’t afford the paid version. The data harvest will be regressive by design.
The robots being trained on this data won’t be free when they arrive. The same companies selling the footage will sell you the robot. You’ll pay for the product that was built from your home’s blueprints.
❓ FAQ
What does Shift do with the data? Shift sells anonymised home footage to robotics and AI companies who use it to train humanoid robots. The company says the data is stripped of identifying information, but the EFF notes that “anonymised” data is often re-identifiable.
Is this legal? In the US, yes — participants consent to the recording. In NZ, the Privacy Act 2020 would require informed consent, but whether “informed” covers the full downstream commercial use of the data is untested.
Could this happen in NZ? Technically nothing stops it. The Privacy Act requires consent, but a company offering free cleaning in exchange for camera access could meet that bar with a terms-of-service checkbox. Whether the Privacy Commissioner would consider that sufficient is unknown.
What’s the actual value of the data? Shift doesn’t disclose pricing. But if you compare it to the cost of a robotics engineer collecting similar data in a lab setting — a single hour of structured home recording could be worth hundreds of dollars to a robotics company. The consumer gets a $150 clean.
