Siri Recordings Are Heard And Reviewed By Apple Contractors

The Guardian:

Apple contractors regularly hear confidential medical information, drug deals, and recordings of couples having sex, as part of their job providing quality control, or “grading”, the company’s Siri voice assistant, the Guardian has learned.

Although Apple does not explicitly disclose it in its consumer-facing privacy documentation, a small proportion of Siri recordings are passed on to contractors working for the company around the world. They are tasked with grading the responses on a variety of factors, including 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.

When your company is running marketing campaigns that constantly accentuate a privacy focus, it is incongruent to keep stuff like this in the shadows. Apple didn’t try to hide that this is happening but it didn’t do much to keep people informed either.

On its machine learning blog, Apple explained how they use customer voice samples to validate the accuracy of "Hey Siri" detection.

The iOS Security white paper says Siri voice recordings are saved for six months with an ‘anonymous’ identifier that can group recordings from the same person together, and then the recordings are kept for at least another two years without the associated identifier. It says the saved audio is “for use by Apple in improving and developing Siri” with “ongoing improvement and quality assurance”. What it does not say explicitly is that these recordings are reviewed and assessed by humans.

If you give it a second of thought, then of course these clips have to be peer-assessed. To improve software to be more human-like, you need humans to label the training data sets and check sample outputs. However, this indirection may not be immediately obvious to people who don’t understand have a firm grasp of how machine learning works. The onus should not be on the customer to know, or to guess. The paragraph should say it explicitly; something like ‘anonymised recordings are reviewed by Apple employees for ongoing improvement and quality assurance of the Siri service’.

It’s also important to point out that the content of the white paper is not what is shown to a user when they activate Siri on their iPhone or iPad. The “Ask Siri, Dictation & Privacy” screen focuses on how Apple handles your data to process Siri requests in the moment, and the copy quickly glazes over the retention policy details. It does not mention the six-month or two-year periods for example.

What is also weird is that Apple has no opt-out for Siri data retention. You can turn off Siri completely and Apple will delete any data on its servers that can be traced back to your account, but you cannot elect to use Siri and not have Apple keep data to improve the service. Plenty of other iOS features provide this granularity, with dedicated options to “Improve Maps”, “Share iCloud Analytics” and the like. It’s a strange omission that Siri offers no such controls.