What Your Robot Vacuum Can See, and What It Sends Home

A robot vacuum is a camera, a set of sensors, a microphone in many models, and a continuously updated map of your home, driving itself between rooms. None of that is sinister by design: it is how the machine avoids table legs and remembers where the kitchen is. But it does mean the device collects a category of data that no other appliance in the house has, and what happens to that data is a vendor policy rather than a law of nature.

Start with the map, because people consistently underestimate it. A floor plan with room labels, furniture positions, and the times of day the machine runs is a precise description of how you live. Vendors use it for scheduling and no-go zones, and some have explored sharing home layout data with other smart-home platforms. It is worth knowing whether your map is stored locally, in the vendor's cloud, or shared onward, and that answer is usually in the privacy policy rather than the box.

The camera is the part that generates headlines, and the mechanism is worth understanding. Navigation cameras feed obstacle recognition, and that recognition is often trained and sometimes run in the cloud. The widely reported incident where images captured by vacuums, including people in private situations, ended up circulating came from exactly that pipeline: images sent for labelling reached human annotators. It was not a hack; it was the normal operation of a training process that the owners did not know about. Research since has repeatedly found connected vacuums with weak transport security, permissive companion apps, or microphones whose purpose is not disclosed clearly.

None of this makes the category unusable. It makes a few purchase-time questions worth asking. Does the model work fully without a cloud account, and does basic cleaning continue if the internet is down? Can the camera be disabled in software, and does the vendor say what disabling actually stops? Is there a documented way to delete maps and recorded data? Does the device support local-only control through a standard like Matter, or a community firmware alternative, so it keeps working if the vendor loses interest? Is the company still shipping security updates for models a few years old?

For a device already in the house, there are two things worth doing today. Put it on a separate network: the guest network, or an IoT VLAN if your router supports one, so a compromised appliance cannot reach your laptop, your NAS or your backups. Then open the companion app's privacy settings and turn off whatever data sharing is optional, including the image-contribution programmes that are frequently on by default and described as helping improve the product.

The broader point generalises to the rest of the smart home. The questions that matter are not whether a device has a camera, but where its data is processed, how long the vendor will support it, and what still works when the company's servers eventually go away.