Your smart speaker is in a class action lawsuit you did not know about.

What the case actually says

The case is Garner v. Amazon, certified in the Northern District of Illinois on July 18, 2025. The legal theory rests on the Illinois Biometric Information Privacy Act, BIPA, which forbids commercial collection of biometric identifiers without specific written consent. A voiceprint is a biometric identifier under the statute. Amazon’s defense, that the voice command itself is the consent, was rejected at certification. The court held that hitting the wake word does not constitute the kind of informed, specific, written consent the statute requires.

The certified class is residents of Illinois who used an Alexa device between January 2018 and December 2024. Other state-level cases are in motion in Texas, California, and Washington under analogous biometric statutes. The federal multi-district consolidation discussed in the procedural orders is likely. The settlement, when it comes, will run into hundreds of millions of dollars. The settlement will not, by itself, force deletion of the underlying voiceprints. That part has to be requested separately.

What Alexa actually records

The voiceprint is the misunderstood piece. It is not the audio of what you said. It is a numerical fingerprint of how you say it: pitch contour, formant frequencies, cadence, the acoustic signature of your particular vocal tract. Once stored, it identifies you across recordings, across households, and across new devices on the same Amazon account. Amazon uses it primarily to distinguish family members for personalized responses. The same fingerprint, in someone else’s hands, identifies you anywhere a similar voice sample appears. That second-order risk is what we cover in our piece on how thirty seconds of audio is now enough to clone a voice convincingly, where the same voiceprint stored for personalization can later be used as raw material for a convincing voice clone.

Beyond the voiceprint, Alexa retains three other categories. The audio recordings of your commands, by default, with timestamps and device location. The transcripts of those recordings, separately. And the metadata, who triggered the wake word, when, on which device, with what subsequent action. The metadata is the smallest in size and the most useful in court, the same pattern we walked through in smart home in custody battles.

How to check your own exposure

Three steps, in order. None of them require a settlement claim and none of them affect your eligibility if a class action notice arrives later.

1. Pull your voice history

Open the Alexa app. Settings, Alexa Privacy, Review Voice History. Set the date range to the maximum the app shows. Scroll. You will see every command, with the audio playable. The recordings span back to your first device unless you have explicitly purged. For most households this is several years.

Note any recordings that capture conversation in the background, conversations you did not intend to address to Alexa. The wake-word false-positive rate produces these, and they are the recordings that show up in family-court subpoenas with the most damaging effect.

2. Export the data Amazon holds

Amazon’s Request My Data flow lives at amazon.com/gp/privacycentral/dsar/preview.html. Select Alexa, request all categories, submit. The export arrives within thirty days. The download is a ZIP archive of JSON files. The file you want contains the recording manifest with timestamps, device IDs, and links to the audio.

Read the archive before you delete anything. Once you understand what is there, you can make a deletion decision. If anything in the archive is relevant to a foreseeable legal matter, talk to a lawyer first. The same spoliation logic we walked through for deleting evidence vs protecting yourself applies to Alexa data.

3. Set retention to zero going forward

Alexa app, Settings, Alexa Privacy, Manage Your Alexa Data, Choose how long to save recordings, set to Don’t save recordings. This is the simplest setting change Amazon offers. Turn it on now and you stop the accumulation immediately. The existing recordings are unaffected by this setting and need to be deleted separately if you decide to delete them.

If you stay in the Alexa ecosystem

Most households will not throw out their Echo speakers. The class action is about consent and damages, not about the underlying functionality. Three settings reduce the data exposure without giving up the device.

Disable voice recognition profiles. Settings, Your Profile, Voice ID, then remove. This stops Amazon from using voiceprint matching to identify which household member is speaking. Recordings still happen, but the biometric identification layer is disabled on the account.

Disable Amazon Sidewalk. Settings, Account Settings, Amazon Sidewalk, off. Sidewalk is the mesh network that uses your home bandwidth to extend coverage for other Amazon devices nearby. It also pushes telemetry through the network in ways that have not been fully documented.

Audit skill permissions. Many Alexa skills request access to voice data and metadata they do not need. Settings, Skills and Games, audit which skills are enabled and disable any you do not actively use.

If you want a voice assistant without Amazon

Two options that work today. Neither is as polished as Alexa. Both are usable.

Apple HomePod with Siri keeps most processing on-device. The audio that does leave the device for cloud processing is not retained by default and is not used to build a centralized voiceprint database. Apple’s privacy posture on voice is structurally different from Amazon’s, the same pattern that gives iPhone users less exposure to geofence warrants than Android users.

Self-hosted assistants like Home Assistant with the Voice Preview Edition or Mycroft AI run entirely on hardware you own. The setup is more involved and the natural-language understanding is rougher. For households where the smart-home function matters more than the assistant function, the trade-off is acceptable.

Frequently asked questions

Am I eligible if I am not a resident of Illinois?

Not in the certified class. Other state-level cases are pending in Texas, California, and Washington under analogous statutes. The federal multi-district consolidation, if it happens, will widen the eligible class. Watch for notice arriving by email or postal mail at the address tied to your Amazon account.

Will deleting my recordings hurt my class action claim?

No. The claim is for past collection without consent. Amazon already has the records of when collection happened. Deleting the user-facing copy leaves Amazon’s internal collection records untouched, and your standing as a class member is determined by device usage in the qualifying period rather than by the current data state.

Does this apply to Echo Show, Echo Auto, or Alexa on Fire devices?

Yes. The class definition covers any device on which Alexa was used in the qualifying period. The voiceprint capture is platform-agnostic on Amazon’s side. Echo Show adds a video layer that has its own separate retention questions. Echo Auto in a vehicle adds a location layer. The biometric voiceprint is the common element.

What about Google Assistant and Apple Siri?

Google has been sued under similar state biometric statutes and has settled some matters. Apple’s design choice to do most processing on-device gives it a structural advantage in this kind of case. The exposure ranking, from most to least: Amazon Alexa, Google Assistant, Apple Siri. None is exempt. The differences are in scale and architecture.

---

There’s no perfect setup. Anyone selling you perfect is selling fear. The goal is simple: make yourself a harder target than the person next to you.