Microsoft employees listen in on Skype translations and Cortana conversations

Microsoft employees sometimes listen in on some Skype conversations when they use the built-in translation feature. They also listen to Cortana recordings. Microsoft does not say in the privacy terms that that data can be listened to by human employees.

This is according to documents that Motherboard has managed to obtain. In addition to documents, the site also obtained screenshots and audio recordings showing that Microsoft employees are listening in on Skype conversations. Microsoft contractors also listen to voice commands that users give to smart assistant Cortana.

Microsoft uses artificial intelligence to automatically translate conversations, and to interpret commands through Cortana. As with other companies such as Google, Apple and Amazon, in some cases this information needs to be looked at more closely, for example when a translation goes wrong or when the wrong wake word is used for a smart assistant. In the case of Skype’s translation function, employees will hear a piece of audio, including some suggested translations suggested by the artificial intelligence. From this they have to choose which translation best suits the audio fragment, but the employees can also provide their own input.

Microsoft tells Motherboard that the audio clips can only be heard through a secure portal, and that the company is trying to omit as much information as possible that identifies users, such as device numbers. In the terms and conditions, the company only states that it “collects and uses conversations” to improve Microsoft’s services, and that that information is “analyzed,” but does not say that human users are involved.

Several major tech manufacturers have come under fire in recent weeks for allowing human employees to listen in to parts of conversations conducted through smart assistants such as Assistant, Siri and Alexa. Google and Apple have recently temporarily stopped collecting, and together with Amazon are offering opt-outs for collecting and analyzing audio recordings.