Amazon staff in India and Romania are employed to watch select snippets of footage from customers Cloud Cam home security cameras, according to Bloomberg.
The ecommerce giant has become embroiled in another privacy scandal just months after reports emerged that Amazon’s teams record and transcribe commands given to Alexa devices in users’ homes.
According to five sources who have worked on the Cloud Cam programme, dozens of staff watch video snippets from inside people’s homes in order to help improve the devices algorithms.
Its Cloud Cam devices alert users to movement and sounds in their homes while they’re away, enabling customers to view the feed remotely via an app.
These teams reportedly watch and annotate around 150 20-30 second videos every day, aiming to train the algorithms to better differentiate between credible threats and false alarms.
An Amazon spokesperson stated that the review clips only come from employee testers and Amazon customers who send in clips for troubleshooting, adding that Amazon takes “privacy seriously and put Cloud Cam customers in control of their video clips”.
However, two of the five sources said the teams have picked up clips of people’s homes they’re unlikely to want to be shared, including instances of people having sex.
These clips are flagged and scrapped so as to improve the experience of the human reviewers according to the spokesperson, though they failed to explain why they were included in voluntarily submitted clips.
Three of the sources added that they don’t know how Amazon selects the clips to be annotated, but that there were no obvious technical glitches that would require submitting for troubleshooting.
It is not uncommon for tech companies to use human trainers to improve their AI algorithms, with reports of both Google and Apple running similar programmes emerging after news of Amazon’s operation broke.
Google and Apple subsequently paused their human review programmes, while Amazon began allowing users to exclude their recordings from manual review.