Google is spying on your voice

In a prior article I wrote how big tech giants are being accused of ambient espionage, and this is yet another piece that proves the legitimacy of these claims.

People were not saying “OK Google.”

They were not asking a question.

The device just… activated.

Google calls them “false accepts.” The lawsuit calls them unauthorized recordings.

The final approval hearing was on March 19, 2026. Payouts should range from $2 to $56 per user.

This is the second major voice assistant settlement in three months. Apple paid $95 million in January for the same thing with Siri — activating without “Hey Siri” and capturing private conversations. Two plaintiffs said they got targeted ads for products they’d only discussed verbally.

Here’s what both settlements confirm:

Voice assistants mishear trigger words. This isn’t a bug — it’s a known architectural tradeoff. The mic is always passively listening. When it incorrectly activates, it records. When it records, that audio goes to a server. And sometimes, humans listen to it. Apple, Google, and Amazon have all acknowledged that contractors review voice samples.

Apple’s review program is now opt-in. It wasn’t always.

The structural question isn’t whether these companies are malicious. It’s whether “cloud by default” is the right architecture for a microphone that’s always on.

$163 million in settlements says the market is answering that question.

If your work involves anything you wouldn’t want on someone else’s server — client conversations, deal terms, personnel decisions — the question isn’t whether your assistant is listening. It’s what happens when it does.

I’m Ari Nakos. I build privisay — local, private voice-to-text for Mac. privisay.com