Apple Contractors Listen to Siri Audio Samples Too

Earlier this month, reports came out that Google’s contractors spread around the world had access to Google Assistant voice recordings. However, it turns out that Apple — the company that claims it’s the best in terms of user-privacy — is no better.

This frequently includes confidential information, such as medical history, sexual interactions, and even drug deals, a whistleblower working for one of the contractors is cited to say. Apple has responded to the report, confirming that a small portion of Siri recordings is indeed used for improvements.

That’s incredibly horrifying, but what’s more worrying is the fact that most of these recordings where confidential and extremely private things are happening, are recorded by Siri because of accidental wake-ups even when users have not used the ‘Hey Siri’ wake word. To make matters worse, these clips have identifying characteristics such as the user’s location, contact details, and app data, according to a whistleblower who spoke to the Guardian on the condition of anonymity.

The Guardian cites a whistleblower at one of the contractors allegedly working for Apple to claim the Cupertino-headquartered company releases a small proportion of Siri recordings to such contractors. These contractors are expected to grade the responses on numerous factors, such as "whether the activation of the voice assistant was deliberate or accidental, whether the query was something Siri could be expected to help with and whether Siri's response was appropriate."

Accidental activations of Siri, where the voice assistant mistakenly hears its wake word, are often fraught with confidential information, the whistleblower adds.

"There have been countless instances of recordings featuring private discussions between doctors and patients, business deals, seemingly criminal dealings, sexual encounters, and so on. These recordings are accompanied by user data showing location, contact details, and app data," the whistleblower is quoted to say.

While Siri is most often associated with iPhone and Mac devices, the contractor claims the Apple Watch and HomePod are in fact the most common sources of accidental activations.

"The regularity of accidental triggers on the watch is incredibly high. The watch can record some snippets that will be 30 seconds - not that long but you can gather a good idea of what's going on," the whistleblower adds.

Staff are encouraged to treat recordings of accidental activations as a "technical problem", but no procedure was said to be in place to deal with sensitive information. The contractor alleges that employees are expected to hit targets as fast as possible. The report adds that the whistleblower's motivation for disclosure were based on fears of such data being misuses, as there purportedly is not much vetting on who works with the data, a high turnover rate of employees, no proper guidelines about privacy, and the possibility to identify the users.

"It wouldn't be difficult to identify the person that you're listening to, especially with accidental triggers - addresses, names and so on," the whistleblower added.

Finally, the report claims Apple doesn't explicitly mention Siri recordings are made available to humans, not just those that directly work for it but even contractors. The recordings are said to be made available with pseudonymised identifiers. The whistleblower emphasises that the company should especially remove the patently false "I only listen when you are talking to me" Siri response to the query "Are you always listening?".

Apple, for its part, said (in what sounds exactly the same thing that Google said), that the data “is used to help Siri and dictation … understand you better and recognise what you say.” and yet, Siri still sucks.

It adds, "A small portion of Siri requests are analysed to improve Siri and dictation. User requests are not associated with the user's Apple ID. Siri responses are analysed in secure facilities and all reviewers are under the obligation to adhere to Apple's strict confidentiality requirements." The Cupertino company is also cited to say that less than 1 percent of daily Siri activations, and only a random subset, are used for grading. These recordings are usually only a few seconds long, the company is reported to add.