Germany Is Investigating Facebook for Transcribing Users' Audio

Facebook is being probed by Hamburg’s data protection authority over transcribing audio from users of its services, adding to an investigation into Google’s automatic speech assistant.

Facebook “is currently the subject of a separate investigation” into transcription of human-to-machine and human-to-human communications, the Hamburg Commissioner for Data Protection said in a press release on Monday. “Manual evaluation was used in Facebook Messenger to optimize the transcription function.”

Bloomberg reported earlier in August that Facebook has been transcribing the audio of users who chose the option in Facebook’s Messenger app to have their voice chats transcribed. The human review was aimed at checking whether Facebook’s artificial intelligence correctly interpreted the messages.

In an emailed statement, the Irish data protection commissioner’s office, Facebook’s main privacy watchdog in the EU, said that while the voice-to-text feature was not offered to users in Europe, Facebook inadvertently manually transcribed the audio clips of fewer than 50 European users across 14 countries. The situations occurred when a U.S.-based user of the feature engaged with a Europe-based contact.

The Irish watchdog said it had notified the relevant national privacy authorities of the 14 countries and said Facebook had pledged to inform it before starting manual transcription of European users’ audio data. It’s unclear how many U.S.-based users were impacted.

European privacy authorities have the power to fine companies as much as 4% of annual turnover for violating the bloc’s data protection rules, but such tough measures are usually reserved for major breaches.

Facebook is facing intense regulatory scrutiny of its businesses in Europe, including an antitrust probe into its Libra cryptocurrency and numerous privacy investigations that could lead to hefty fines.

Facebook declined to comment. The company previously said it had paused human review of the audio in early August.

Google said in a statement that it has met with the Hamburg regulator to discuss the issue.

“We’re currently assessing how we conduct audio reviews and help our users understand how data is used,” Google said. “These reviews help make voice recognition systems more inclusive of different accents and dialects across languages.”

The issue of digital assistants and transcription of user commands has dogged several major technology companies. Tech platforms say the human analysis of voice chats helps train and correct mistakes, thereby improving responses to queries by users. But privacy advocates worry users’ rights could be breached if terms of service about the practice are unclear or transcription occurs without their knowledge.

Hamburg has been probing Google over the recordings and the company has already agreed to pause the practice for at least three months. The Alphabet unit should warn users about the risk of accidental activation of speech recordings and seek consent for transcriptions, the regulator said.

Apple and Amazon have also suspended or announced changes to human analysis of user commands, amid increasing global scrutiny by regulators and lawmakers over the privacy implications of the practice. The U.S. Congress is working on a federal privacy bill that could also tackle the handling of voice recordings.

More recently, Microsoft was found to engage in the practice, according to articles by Motherboard, highlighting the importance human review still plays in training machine-learning algorithms.

In a separate regulatory case Monday, Facebook won temporary suspension of a German antitrust order banning it from combining user data of various services. The Federal Cartel office, which issued the order, said it will appeal.