The Paris prosecutor’s office has opened an investigation into Apple for alleged illicit data collection via its voice assistant Siri. The case has been entrusted to the Cybercrime Unit.
This proceeding follows a complaint filed by the League for the Defense of Human Rights (LDH), Politico recalls, which revealed the action of the Prosecutor’s Office. The complaint rests on the testimony of Thomas Le Bonniec, a technology researcher and former Apple subcontractor in Ireland, who confirmed that he originated the report.
According to the complaint, Apple would have collected, recorded and analyzed conversations through Siri without users’ consent. Thomas Le Bonniec publicly spoke about his experience analyzing sensitive user recordings, including conversations of cancer patients.
Apple Denies Data Sharing with Third Parties
The Siri voice assistant, present on most Apple devices, can record and retain audio interactions to improve its services. These data can be kept for up to two years and reviewed by “graders,” contractors responsible for evaluating the quality of the responses.
Apple points to one of its publications (January 2025) stating that conversations with Siri are never shared with marketing specialists nor sold to advertisers. And it clarifies that it “does not retain audio recordings of interactions with Siri unless users explicitly agree to help improve Siri, and even then, the recordings are used solely for that purpose.”
Testified in June last year before the Senate Economic Affairs Committee, Renault’s Chief Scientific Officer and co-designer of Siri, Luc Julia, also rejected the possibility of eavesdropping. “Siri does not listen to us […] It’s physically impossible, in fact. Well, I’m quite well positioned to tell you. It’s absolutely impossible to listen to everything, all the time and to store it on disks. It’s not possible. It’s silly to think that would be possible. On the other hand, the attacks that are currently being leveled at Siri are attacks that make sense because sometimes there are phrases that are not properly recognized and we consider that there is a limit. This limit is at 70% and when we are not sure, less than 70% of the recognition of the phrase; it is extracted, anonymized and sent to humans who will listen to the phrase to know why it was not recognized correctly and how we can improve the model so that next time a phrase of the same type is better recognized. So it’s rare. It’s very rare compared with the 500 million daily users of Siri today.”
