Google Assistant-enabled devices allegedly found to listen in on users
VRT NWS has released a report in which it claims to have acquired audio recordings made by Google Assistant in a surreptitious manner without the knowledge of the user in question. The Belgian news outfit apparently matched one of these recordings to specific individuals located in the muncipality of Waasmunster. It then tracked this couple down, one of whom affirmed that he could recognize the voices of his relatives in the applicable audio clips.
This material was disclosed to VRT by a source claiming to be a Google subcontractor whose job it was to validate speech-recognition transcripts through audio descriptions. This task involves the use of Crowdsource (an online resource that allows people to contribute to the company's search algorithms through human interpretations of various communication methods). In the subcontractors' case, they have access to a secure form of this tool so as to improve language analysis at Google using real-life excerpts of recordings made via the Assistant system.
VRT NWS has also asserted that these workers may be allowed access to private information, including the disclosure of sensitive information and interpersonal interactions, that is captured by the Assistant during these recordings. Furthermore, it appears that some recordings are made following the inadvertent use of, or perception of, of a 'wake word' or other form of activation.
Should this be the case, it is possible that Google is in violation of the European Union's GDPR regulations in some cases. The Mountain View company rationalizes these practices by declaring that it needs them in order to help its AI learn fine distinctions and variations while listening to languages such as Dutch.
Furthermore, it has claimed that only 0.2% of these recordings are used for this purpose. However, should VRT NWS' findings be valid, this is still a potential source of concern for those who would like to preserve their privacy while using Google Assistant.