Which companies paid settlements over voice recordings?

Three large technology companies have settled over voice recordings. Apple agreed to pay $95 million million US dollars ¹ in a proposed class action over Siri. Google agreed to pay $68 million million US dollars ² over Google Assistant. Amazon agreed to a 25 million US dollar civil penalty ³ over Alexa and children's voice data. Everything below comes from our dataset of 13 data points ⁴, drawn on 13 sources ⁵.

Voice assistants drew the three biggest payments

Each of the three large payments is tied to a voice assistant that was meant to listen only after a wake word.

• Apple, Siri. Reuters reported in January 2025 that Apple agreed to pay a cash settlement of $95 million million US dollars ¹. The claim was that Siri violated users' privacy by recording without consent. The settlement was proposed and needed court approval.
• Google, Google Assistant. In January 2026, Google agreed to pay $68 million million US dollars ² in a class action in federal court in San Jose. Google denied wrongdoing.
• Amazon, Alexa. The FTC and the Department of Justice announced a permanent injunction and a 25 million US dollar civil penalty ³ over alleged violations of the children's privacy law COPPA. The Associated Press reported the same figure ⁶. Amazon said it disagreed with the claims and denied violating the law. The two entries are one settlement, reported by two sources.

Smaller cases show the same issues at other scales

The table lists rows from the dataset. Rows without an amount (investigations, filed lawsuits and orders) are discussed below it.

Showing 8 of 13 rows. See every row in the dataset.

Other payments are much smaller but concern the same behavior:

• Whole Foods reached a settlement of nearly 300000 million US dollars ¹³ in Illinois. Plaintiffs said headsets collected warehouse workers' voiceprints under the state's biometric privacy law. A court gave preliminary approval in January 2023.
• Cox Media Group agreed to pay $880,000 million US dollars ¹⁴ to settle FTC charges that it falsely claimed to offer "active listening" ad targeting to small businesses. The issue is a false claim about listening, not proven recording.
• Spain's data protection agency fined an unnamed company a few thousand euros for recording a customer's call without warning or consent ¹⁰.

Some of the biggest cases have no clean voice amount

Several rows carry text instead of a number, on purpose.

• Google and the Texas Attorney General. In May 2025 Google agreed to a settlement of Texas privacy claims, and the Attorney General alleged that Nest and Google Assistant captured voiceprints ¹⁵. The sum also covers a separate location-data suit, and the split is not stated. It is not a voice-only figure, so the dataset records it as text, not as an amount.
• Kenya. The Data Protection Commissioner ordered Liquid Telecommunications Kenya to compensate a complainant in Kenyan shillings for recording an exit consultation call without consent ⁸. It is compensation to one person, in shillings, not a fine.
• Italy. The Garante declared unlawful the processing of call center recordings in 2012 ¹⁶. It was an order, with no amount.
• Otter.ai. A consolidated class action over the Otter Notetaker meeting assistant was filed in federal court in California ⁹. No settlement or damages amount has been reported.

Investigations are also on file. The Hamburg data protection authority said in 2019 that having contractors or staff review voice assistant audio without additional informed consent violates users' rights ¹¹. In 2025 the Paris prosecutor's office opened an investigation into Siri ¹², after a complaint by the Ligue des droits de l'homme ⁷.

The chart compares size, and size is not the whole story

The chart ranks the largest voice privacy settlements and fines by amount. A bigger number does not mean worse behavior. It reflects the number of users in a class, the law used and the country.

What these numbers do and do not measure

• A settlement is not a finding. Companies in these cases generally denied wrongdoing, and many rows report allegations only.
• Some amounts were not final. The Apple figure was a proposal needing court approval. The Google 2026 figure is described as filed, so check the docket before saying a judge approved it. The Whole Foods figure is a preliminary approval.
• Amounts are as reported. Several come from news reports, not the regulator's own page. The Spanish fine names no company and gives no decision date, so its year is the report year.
• Only published cases appear. The dataset holds what we could find and quote. Confidential settlements, and cases with no stated amount, are missing or recorded as text.
• Rows are not comparable in kind. They mix class action settlements, civil penalties, one complainant's compensation and unfinished investigations.

What this means for choosing how to dictate

The payments cluster around three problems: recording when the user did not expect it, humans reviewing recordings, and collecting voiceprints or children's voice data. Those are the questions to ask of any tool.

1. Where does the audio go? On-device processing keeps recordings off a vendor's servers. Yap, our own dictation app for Mac and Windows, works on-device, and our on-device dictation guide explains what that means. Other tools may fit you better, and you can compare them on our market page.
2. Does anyone listen to samples? Read the vendor's policy on human review. Our guide to human review of voice recordings explains what to look for.
3. Is the tool always listening, or only when you press a key? Push to talk limits what can be recorded. Meeting assistants that join calls raise the consent question for everyone else on the call, as the Otter.ai case shows.
4. Who else is in the room? Consent rules differ by country and state, so ask other people before recording them.

For the wider picture, see our overview of voice privacy and the full dataset, which lists every row with its quote and source.