Facebook’s Messenger app – the second most-popular messaging app after Facebook-owned WhatsApp – allows users to exchange written messages, images and other media, including voice recordings and transcriptions of voice recordings. The latter ‘Voice to Text’ feature was introduced in 2015.

According to Bloomberg, Facebook used contract workers from companies including regular client TaskUs Inc to review clips recorded and sent in Facebook Messenger. The employees were not told where the audio was recorded, how it was obtained, or why they needed to perform the work, only to transcribe it. The workers told Bloomberg that they sometimes heard ‘vulgar content’.

In its recent data use policy, Facebook does not disclose to users that third parties might review their audio recordings.

Facebook confirmed that it employed workers to review some samples of audio in order to improve its natural-language processing capabilities. Facebook added that the audio used in the programme was anonymised to grant users a degree of privacy and that the company never listened to audio recordings without being granted permission.

However, according to Facebook’s support page, if even one person in a group chat consents to Facebook transcribing the conversation, any audio may be transcribed and passed on for human review.

The company stated that it has now ended the programme: “Much like Apple and Google, we paused human review of audio more than a week ago,” a Facebook spokesperson said.

Facebook has been the subject of numerous privacy scandals in the past two years. Recently, the company agreed to pay a £4bn fine to the US Federal Trade Commission over its role in the Cambridge Analytica scandal, in which a data analytics company harvested data from 87 million Facebook users to build psychologically targeted political adverts.

Already this year, a clutch of large tech companies have been criticised for using employees and contractors to listen to voice recordings of people speaking to their virtual assistants. In April, Bloomberg revealed that thousands of workers were listening to clips of people speaking to its Alexa assistant to improve its natural-language processing capabilities. These clips included recordings taken by accident, as well as some featuring embarrassing or potentially criminal activities. Amazon has since allowed customers to prevent humans listening to their voice commands, but has not put a stop entirely to the practice.

A Guardian report in July revealed that Apple had employed contractors to listen to recordings of people speaking to the Siri assistant, with some recordings including highly sensitive data (such as medical information or the sounds of people having sex). Apple has since suspended the programme globally. Google runs a similar programme to improve the capabilities of Google Assistant and recently agreed to pause the programme for three months in the EU while an investigation by German regulators is carried out.

More recently, Motherboard revealed that Microsoft had also paid contractors to manually review voice recordings from Skype and its Cortana virtual assistant. One contractor commented that they heard people discussing relationship issues, searching for pornography, revealing their full addresses and having phone sex.

Facebook’s Messenger app – the second most-popular messaging app after Facebook-owned WhatsApp – allows users to exchange written messages, images and other media, including voice recordings and transcriptions of voice recordings. The latter ‘Voice to Text’ feature was introduced in 2015.

According to Bloomberg, Facebook used contract workers from companies including regular client TaskUs Inc to review clips recorded and sent in Facebook Messenger. The employees were not told where the audio was recorded, how it was obtained, or why they needed to perform the work, only to transcribe it. The workers told Bloomberg that they sometimes heard ‘vulgar content’.

In its recent data use policy, Facebook does not disclose to users that third parties might review their audio recordings.

Facebook confirmed that it employed workers to review some samples of audio in order to improve its natural-language processing capabilities. Facebook added that the audio used in the programme was anonymised to grant users a degree of privacy and that the company never listened to audio recordings without being granted permission.

However, according to Facebook’s support page, if even one person in a group chat consents to Facebook transcribing the conversation, any audio may be transcribed and passed on for human review.

The company stated that it has now ended the programme: “Much like Apple and Google, we paused human review of audio more than a week ago,” a Facebook spokesperson said.

Facebook has been the subject of numerous privacy scandals in the past two years. Recently, the company agreed to pay a £4bn fine to the US Federal Trade Commission over its role in the Cambridge Analytica scandal, in which a data analytics company harvested data from 87 million Facebook users to build psychologically targeted political adverts.

Already this year, a clutch of large tech companies have been criticised for using employees and contractors to listen to voice recordings of people speaking to their virtual assistants. In April, Bloomberg revealed that thousands of workers were listening to clips of people speaking to its Alexa assistant to improve its natural-language processing capabilities. These clips included recordings taken by accident, as well as some featuring embarrassing or potentially criminal activities. Amazon has since allowed customers to prevent humans listening to their voice commands, but has not put a stop entirely to the practice.

A Guardian report in July revealed that Apple had employed contractors to listen to recordings of people speaking to the Siri assistant, with some recordings including highly sensitive data (such as medical information or the sounds of people having sex). Apple has since suspended the programme globally. Google runs a similar programme to improve the capabilities of Google Assistant and recently agreed to pause the programme for three months in the EU while an investigation by German regulators is carried out.

More recently, Motherboard revealed that Microsoft had also paid contractors to manually review voice recordings from Skype and its Cortana virtual assistant. One contractor commented that they heard people discussing relationship issues, searching for pornography, revealing their full addresses and having phone sex.