Tech companies can do a lot more to protect the identities of people speaking in recordings used to train AI, like shifting the voice or gender of the speaker
It turns out that Siri, Alexa, and whatever it is you call Facebook Messenger have been a little loose-lipped with your conversations.
The piece's contribution is that the fix is technical, not just policy: shifting a speaker's voice or gender before clips reach reviewers would decouple model improvement from identity exposure. By December, Bloomberg's investigation of Amazon, Apple, Google, and Facebook confirmed the industry was still leaning on contractors with raw audio.
First-order effects
Contractors at Amazon, Apple, Google, and Facebook reviewing voice clips would lose access to identifiable speech if voice- or gender-shifting were applied upstream, directly shrinking the exposure documented in Alexa's training process.
Siri, Alexa, and Facebook Messenger owners face an immediate choice between adopting anonymization in their labeling pipelines and defending the current practice publicly.
Second-order effects
If one assistant maker ships voice-shifted training audio first, it turns reviewer privacy into a competitive claim the others must match, pressuring the shared contractor ecosystem that all four rely on.
Anonymized training corpora also cut both ways for users: the same voice-morphing capability that protects speakers is adjacent to the voice-cloning tools later shown defeating bank biometrics, sharpening scrutiny of who controls synthetic-voice technology.
Third-order effects
If the pattern holds, 'de-identify the training data' becomes a baseline expectation for any product that records humans — moving voice-assistant privacy from consent disclosures toward engineering requirements regulators can audit.
Assistants whose value depends on ambient listening would then compete on pipeline design rather than features alone, since the recordings feeding models are the liability surface.
The trend: Voice-assistant privacy is shifting from disclosure-and-consent debates toward technical anonymization of the recordings themselves as the default input to AI training.
Tech companies want to improve Siri, Alexa, and Google Assistant. Can they do it without listening to what we say around our devices? https://slate.com/...
New: after we found Microsoft is hiring contractors to listen to some Skype calls, the company has updated its privacy policy and other pages to explicitly say humans/employees may listen to audio. Wasn't clear before. Original piece based on leaked docs https://www.vice.com/... …
NEW: Leaked documents from a Microsoft contractor show that the humans who listen to & transcribe your conversations have horrible, terrible, underpaid jobs http://www.vice.com/... https://twitter.com/...
These parts of the training materials for contractors listening to Cortana recordings are... a bit on the nose https://www.vice.com/... https://twitter.com/...
All the big tech co's are hiring real humans to do the work of making sure their voice assistants are useful. How much do these contractors, who are left to listen to random users' voice conversations, get paid? Very very little! https://www.vice.com/...
New: leaked docs show what it's really like behind scenes of a tech giant's artificial intelligence system. For Microsoft's Cortana, contractors paid $12-14/hour, with extra $1 if they hit a milestone. Training manuals of 100s of pages of monotonous tasks. https://www.vice.com/..…
Increasingly, every tech company is revealed to be the fucking train from Snowpiercer. Rip off the lid and it's not breakthrough technology, it's abusive labor practices. https://twitter.com/...
Microsoft's privacy statement now says “Our processing of personal data for these purposes includes both automated and *manual (human) methods* of processing” While Facebook, Apple, Google have suspended use of humans though, Microsoft holding on https://www.vice.com/... https://…
Facebook confirms it has done this, but says it's not happening anymore: “Much like Apple and Google, we paused human review of audio more than a week ago.”