Source and documents show Microsoft contractors are listening to some Skype calls made using the app's translation service
Contractors working for Microsoft are listening to personal conversations of Skype users conducted through the app's translation service, according to a cache of internal documents …
Context & Ripple Effects
The Skype Translator that Microsoft baked into Windows PCs in 2015 turns out to have had a human layer behind it: internal documents show contractors listened to personal calls routed through the translation service. The disclosure lands mid-scandal — days after VICE reported Microsoft contractors heard accidentally invoked Xbox recordings, many from children, and a week before Microsoft said it had already stopped reviewing Xbox audio for product improvement.
The pattern is industry-wide, not Microsoft-specific: weeks earlier Google admitted one of its language reviewers leaked confidential Dutch audio. What distinguishes Microsoft's exposure is scale and economics — leaked documents show Cortana transcription contractors are paid $12–$14 an hour to process up to 200 clips per hour, and a later report found Skype audio vetting was done by contractors working from home in China with no security measures.
First-order effects
- Skype users who assumed translation was machine-only learn their conversations were human-reviewed without clear consent, putting Microsoft's privacy disclosures for Translator and Cortana under immediate scrutiny alongside its Xbox practices.
Second-order effects
- Microsoft is forced into defensive rollouts across every voice surface at once — it publicly states it stopped listening to Xbox recordings months ago while conceding it still reviews user-to-user audio on TOS-violation reports, narrowing what 'we don't listen' means.
- Rivals face the same disclosure pressure: Google's Dutch audio leak shows the contractor-review model is shared, so any one company's leak raises consent and handling questions for the whole voice-assistant category.
Third-order effects
- If the pattern holds, voice data pipelines get restructured around where and how humans touch audio — the later finding that Skype vetting ran from unsecured home offices in China points toward centralized secure facilities and audited contractor chains as the price of keeping human-in-the-loop training viable.
- Regulators and users converge on treating recorded speech as sensitive personal data by default, forcing explicit opt-in labeling for any feature that routes consumer audio through human reviewers.
The trend: Consumer voice features sold as AI are revealing a hidden human-review workforce, pushing Microsoft and its rivals toward disclosed consent, secured facilities, and tighter custody over recorded speech.