Anker, the company that makes Eufy security cameras, has been paying customers for data in the form of surveillance clips to help train its AI — an approach that underscores just how hungry smart home brands are for labeled real-world data and just how messy their hunger can get when it comes to privacy.
The company offered Eufy users the ability to upload videos of package thefts and people testing car doors, encouraging staged scenarios, it said, and paid a small fee for each qualifying video.
The goal, according to company materials, is to make automatic detection more precise so cameras are better at flagging suspicious behavior without relying entirely on cloud processing.
How the Eufy video training program actually worked
Eufy’s request for footage framed certain goals: tens of thousands of examples for two categories — package theft and car door pulling — to support supervised learning pipelines. Owners were encouraged to submit both real incidents and reenactments, with a $2 bounty per accepted video designed to produce volume fast.
Users submitted clips through a form along with payment information, often using a consumer payments system. Based on comments left by users on the announcement page, over a hundred owners participated — an unimpressive number perhaps, but telling evidence that some portion of consumers will exchange home security footage for cash when the terms seem clear and the task is easy.
Technically, very specific labeled clips — “porch pirate takes parcel,” “person tugs on car handle” — are gold for training detectors and preventing false alarms.
The wager is that a small per-video incentive may yield tons of the same kinds of corner-case content relatively fast, leapfrogging slower, purely organic data collection in the process.
From Cash to Bounties to Video Donations
Along with the paid push, Eufy has been promoting a video donation in-app program, exchanging cash for status and giveaways. Users whose clips contain people could earn gamified badges — an “Apprentice Medal,” for example — and be eligible for prizes like gift cards or hardware.
An “Honor Wall” lists high-volume donors, with the top account claiming more than 200,000 submitted events — a jaw-dropping number that vouchsafes how much footage can gush when contribution is just a tap away. Eufy claims videos donated are used to train and enhance its AI but are not shared with third parties.
The company has made similar requests for baby monitor clips, which are an especially sensitive category — though that, too, does not seem to result in a direct financial payoff. That tension — asking for intimate home recordings to train its model while placing a heavy emphasis on internal use — is at the core of the argument.
Privacy and consent questions raised by the program
Bystanders, delivery workers, neighbors, license plates and children are all bound to be caught on security cameras. Even scripted performances can inadvertently capture faces or voices on adjacent sidewalks or in neighboring apartments. Clear notice and consent for anyone other than the device owner can be difficult to ensure, raising questions highlighted by privacy advocates and consumer groups.
Eufy’s trust record didn’t help, either: Previous reporting by The Verge uncovered that streams from the web portal did not align with its advertised end-to-end encryption, leading Anker to admit there were holes and promise a fix. But when a brand later asks users to send in massive quantities of footage, earlier missteps are not likely to be forgotten.
Regulatory and industry context for AI video training
Regulators have signaled growing scrutiny. The Federal Trade Commission has even fined companies for their handling of home video and, in at least one case, limited how much customer footage could be used to train algorithms without explicit consent. In Europe, the GDPR’s core principles of data minimization and lawful basis apply to biometric and personally identifiable video first and foremost. California’s privacy law imposes disclosure and opt-out requirements that can make widespread AI training programs cumbersome.
Throughout the industry, the playbook is changing. Clips of driver-assistance collected by automakers, data companies from doorbell camera networks that collect incidents and smart appliance makers logging in-home activity have all experimented with opt-in contributions and incentives. The trend reflects a simple logic: modern AI thrives on vast amounts of diverse, well-labeled examples — which is exactly the kind consumers can provide.
Data quality and bias trade-offs in staged incidents
When you pay for staged incidents, you can expedite data gathering but run the risk of biasing models towards dramatic behaviors that do not reflect real-world theft. Should contributors use the same angles and motions ad infinitum to achieve maximum payouts, models might overfit to those patterns, ignoring subtler cues found in the wild. Quality control, environmental diversity, and strict validation sets are the means to address such effects.
There are also some safer design choices: on-device learning so that raw video stays local, automated blurring for faces and plates, and differential privacy to limit the risk of re-identification. Despite stating that videos contributed to the company are only used to enhance its AI and not shared with third parties, Eufy does not publicly detail de-identification or retention practices experts tend to seek.
What this means for smart camera owners and buyers
Anticipate additional calls for “donations” or monetization of footage as brands rush to enhance their sensors. It’s a simple trade-off: some amount of dollars or badges in exchange for the expansion of your home videos’ permanent footprint inside an AI training corpus. Savvy owners will seek out clear consent flows that let users review and revoke submissions, as well as proof that some kind of privacy measures are in place behind the marketing speak.
For Anker, the strategic gamble is that clear policies, trustable security and quantifiable improvements in reducing false alarms and event detection will convince users to keep giving up their data. Without that, it risks landing on the radar of watchdogs — and a consumer base that more and more often is equating “smarter” with “more sensitive” when it comes to the cameras monitoring their front doors and nurseries.
