Voice Recording Privacy Settlements and Fines

Voice privacy money is concentrated in a few large United States cases against the makers of voice assistants. Apple agreed to pay $95 million million US dollars ¹ over Siri, Google agreed to pay $68 million million US dollars ² over Google Assistant, and Amazon agreed to a 25 million US dollar ³ civil penalty over Alexa and children. Everything else in the dataset is either far smaller or has no amount at all. The dataset holds 13 data points from 13 sources, so it is a record of notable cases, not a census.

The reasons differ by who is acting. Private plaintiffs in the United States sue over recording without consent and win large sums. Regulators in Europe tend to state a legal position, open an inquiry or fine small amounts. The sections below set out that split, where the figures are weaker than they look, and what it means for choosing a way to write by voice.

The biggest payments come from class actions over recording without consent

Two of the three largest amounts are class action settlements, and both concern assistants that allegedly recorded people when they had not meant to speak to the device. Reuters reported that Apple agreed to pay $95 million million US dollars in cash to settle a proposed class action claiming Siri violated users' privacy ¹. Google agreed to pay $68 million million US dollars to settle a class action accusing it of illegally recording private conversations through Google Assistant, and it denied wrongdoing ².

A settlement is not a verdict. Neither company admitted fault in the sources we have, and both rows carry caveats about approval. The Apple settlement was proposed and needed court approval ¹. The Google report describes the agreement as filed, so a judge's approval should be checked on the court docket before anyone states it as final ².

Children's voice data produced the largest regulator penalty

The one large regulator payment is Amazon's. Amazon agreed to a permanent injunction and a 25 million US dollar civil penalty in a settlement with the FTC and the Department of Justice over alleged violations of the child privacy law COPPA and the FTC Act relating to Alexa ³. The Associated Press reported the same figure, $25 million million US dollars, and noted that Amazon disagreed with the FTC's claims and denied violating the law ⁵.

These are two rows for one event: the Justice Department's release and a news report. A reader who adds up the table should count the Amazon penalty once. The duplication is deliberate, because the two sources give different context: one is the government's own page, the other records Amazon's denial.

Children are the group where a regulator had a specific statute to apply. The other regulator rows rely on general data protection or consumer protection law, and their amounts are far lower.

Outside the United States, fines are small and many cases are only investigations

European authorities appear in the data mostly as bodies that state a position or open a case, not as collectors of large sums.

• Spain's data protection agency fined an unnamed company 5,000 million US dollars ⁶ for recording a customer's call without warning or consent. The source reported the fine in euros, and the dataset stores the converted amount.
• Hamburg's data protection authority said that contractors or staff listening to voice assistant audio without additional informed consent violates users' rights ⁷. That row has no amount and is a statement of legal position, not a fine.
• The Paris prosecutor's office opened an investigation into Siri ⁸, after a complaint from the Ligue des droits de l'homme alleging non-consensual recordings ⁹.
• Italy's data protection authority declared unlawful the processing of a call center's recorded calls ¹⁰, an order with no amount.

The French rows show how early in the process many cases are. An opened inquiry and a filed complaint are not findings. The Paris investigation article reports an inquiry, not a conclusion that recordings were made without consent or kept, and the prosecutor's own statement was not found on an official page ⁸. The complaint is an advocacy group's allegation, and the report gives no outcome ⁹.

Not every row is about listening devices, and some are about advertising claims

The FTC case against Cox Media Group ended in a payment of $880,000 million US dollars ¹¹, which the source describes as settling charges that the company falsely claimed to offer active listening ad targeting to small businesses. The charge concerns a claim made to customers, not proof that a phone was secretly recording. It is in the dataset because the subject is voice data, but it should not be read as a recording case in the same sense as Siri or Google Assistant.

The Otter.ai case shows the newer front: meeting transcription tools. Otter.ai is the defendant in a consolidated class action over its Otter Notetaker meeting assistant, with the operative complaint dated in December of last year ¹². It has no amount. The complaint's statement that the amount in controversy exceeds a threshold is a jurisdictional figure, not money paid, and we do not use it.

What the data cannot tell you

The amounts are not comparable like for like. A civil penalty, a class settlement and a small regulator fine are different instruments. The class settlements pay many users, the penalty goes to the government, and none of them is a measure of the harm done.

Several rows are allegations or proposals. The Otter.ai complaint states the plaintiffs' allegations, not findings ¹². The later status of Otter's motion to dismiss was not verified. The Apple settlement needed approval when reported ¹.

Some sources are secondhand. The Spanish fine comes from a legal news site, not the agency's resolution, names no company and gives no decision date, so its year is the report year ⁶. The AP figure for Amazon was used because the FTC's own release was not reachable at the time ⁵. News coverage of the Amazon matter also disagrees with the Justice Department's release on the date ³.

Old and new law are mixed. The Italian order applies a privacy code that the GDPR has since replaced, and it is mainly about monitoring workers, with call recording as the means ¹⁰. It tells you little about current enforcement.

The sample is small and leans toward big names. The data covers only cases that reached a reportable stage with a stated source. Many smaller or settled-in-private disputes will be missing, and missing cases are more likely to be small. Treat the amounts as the visible top of the market, not its average.

What this means if you choose how to write by voice

The cases point to three practical questions about any dictation tool, none of which is a legal assessment of any company.

• Is audio kept, and who can listen? The Hamburg statement is about human review of recordings without informed consent ⁷, and the Siri investigation is filed under the same issue ⁸. Check whether a tool sends audio to a server and whether people may review it.
• Is it always listening, or only when you press a key? The large class actions are about recordings the user did not intend. A tool that records only while you hold a key or press a button gives less to dispute.
• Who else is on the recording? The Spanish fine concerned a customer who was not warned ⁶, and the Otter.ai complaint concerns a meeting tool ¹². If you dictate or transcribe meetings and calls, the other people's consent matters, not only yours.

On-device dictation, which Yap offers and which we make, keeps audio on your own computer, so the questions about retention and human review do not arise in the same way. Built-in tools may be enough for you, so compare them by reading each vendor's privacy page. Our overview of voice privacy explains what to look for in those policies.

All rows

Showing 12 of 13 rows. See every row in the dataset.