AI Wearable Recorders and Your Privacy: What Limitless, Plaud, and Bee Actually Do With Your Conversations

The short answer: AI wearable recorders—Limitless, Plaud Note, Bee, and similar pendants—capture continuous audio from every room you enter and every person you talk to, then transcribe and store it on the vendor's cloud by default. Most people wearing one have never audited where those transcripts go, how long they're kept, or whether the people around them consented to being recorded at all. You can use these devices without handing over a permanent, searchable record of your life—but it takes deliberate configuration, not the default install.

AI wearables went from novelty to mainstream accessory faster than almost any hardware category in recent memory. A Limitless pendant or a Plaud Note clipped to a lanyard now shows up in meetings, at dinner, in cars, and on hallway walks that used to be genuinely private. The pitch is compelling: never take notes again, never forget a commitment, search your entire life like you'd search your email. The part that gets less attention is what's actually happening on the vendor's servers while that pitch plays out.

This isn't a guide telling you to throw the device away. It's a guide to using it like someone who understands what continuous audio capture actually creates: a permanent, cloud-hosted, third-party-searchable transcript of everyone who has ever spoken near you.

What These Devices Actually Capture

The category includes a handful of well-known products, and they don't all work the same way—but the shared architecture is what matters for privacy.

Limitless (formerly Rewind Pendant) is a wearable microphone that records ambient audio throughout the day, transcribes it, and builds a searchable timeline of conversations, meetings, and voice memos. It syncs to Limitless's cloud by default, where transcription and the "Limitless memory" search index live.

Plaud Note is a card-sized recorder that pairs with a phone app. It records calls and in-person conversations, then uploads audio to Plaud's servers for AI transcription and summarization. The raw audio and transcript both leave your device.

Bee (and similar pendants like Friend) take ambient capture further—always-on listening designed to build a running model of your habits, relationships, and preferences, marketed explicitly as a proactive AI companion rather than a note-taking tool.

The common thread: audio is captured on a device you control, then transmitted to a server you don't, where a third party transcribes it, stores it, and in most cases uses it to power a product feature (search, summarization, proactive suggestions) that requires the content to remain accessible on their infrastructure indefinitely.

That's a meaningfully different risk profile than a cloud AI chat session. A ChatGPT conversation is something you chose to type. An ambient recorder captures whatever anyone near you says—coworkers discussing a deal, a doctor on a phone call in the next room, your kid's friend venting about their parents—without any of those people knowing a permanent transcript is being generated. It's a sharper version of the always-on tradeoff covered in building a private AI voice assistant—except a wearable pendant doesn't have a local, no-cloud option the way a home voice assistant does.

The Legal Reality: One-Party vs. Two-Party Consent

Before anything technical, there's a legal question that most wearable owners never ask: is it even legal to record the people around you?

The United States splits into two regimes. One-party consent states (the majority, including New York, Texas, and Florida) only require that one participant in the conversation—you—consent to the recording. Two-party (all-party) consent states—California, Illinois, Pennsylvania, Washington, and about a dozen others—require every person in the conversation to consent, or the recording can be illegal regardless of your intent.

If you travel for work, or if the people you talk to travel, an AI pendant that's fine to wear in Texas can create a legal problem in California. The device doesn't know or care what state you're in. You do. This isn't a privacy nicety—unauthorized recording in a two-party state can carry civil liability and, in some states, criminal exposure.

The practical fix is simple: know your state's law, know the states you regularly visit, and default to disclosure ("I'm wearing an AI recorder, is that okay?") when the law requires it or when the conversation is sensitive regardless of the law. Most of these devices have a mute or pause gesture—use it walking into any conversation where consent hasn't been established.

What Happens to the Transcript After It's Created

This is the part vendors bury in privacy policies most users never open. Three questions matter more than any marketing claim:

Is it used for model training? Some AI wearable companies reserve the right to use captured audio and transcripts to improve their models unless you opt out. Read the specific clause—"improve our services" is broad enough to cover training. If there's no explicit opt-out toggle in settings, assume the default is permissive.

How long is it retained? Most of these products default to indefinite retention because the entire value proposition is a searchable lifetime timeline. That means a subpoena, a breach, or an account compromise years from now exposes conversations you've long forgotten having.

Where does it live? Transcripts are typically stored on the vendor's cloud infrastructure (AWS, GCP) under the vendor's encryption keys, not yours. "Encrypted at rest" without zero-knowledge architecture means the vendor—and anyone who compels the vendor—can read the plaintext. Very few AI wearable companies currently offer true end-to-end encryption where even they can't decrypt your transcripts, because it would break their own search and summarization features.

Run the actual privacy policy through a research pass before you buy or keep using one of these devices—don't rely on the marketing page. Perplexity is useful here specifically because it cites primary sources rather than summarizing secondhand blog takes: ask it to pull the current retention and training clauses directly from the vendor's published privacy policy and terms of service, with links, so you're reading the actual language rather than someone's interpretation of it from six months ago (policies change). This is one specific case of a broader habit worth building for any new AI product—see how to vet any AI tool before it touches your data for the full due-diligence checklist.

Affiliate Disclosure: This article may contain affiliate links. If you make a purchase through these links, we may earn a small commission at no extra cost to you. We only recommend products we genuinely believe in. This helps support our work and allows us to continue providing free content.

Building a Privacy-Conscious Wearable Workflow

You don't need to abandon the category to control the exposure. Five changes cover most of the risk.

1. Turn off training opt-in immediately. Every product in this category has a settings toggle for whether your audio can be used to train models. It's frequently on by default. Turn it off the day you set up the device, not after you've generated months of transcripts.

2. Set a retention limit, and actually use it. If the app offers auto-delete after 30, 60, or 90 days, set it. A rolling window still gives you the note-taking benefit without accumulating a multi-year archive that outlives your reason for keeping it.

3. Export anything you actually need, then delete the vendor copy. For the handful of conversations you genuinely want to keep—a specific meeting, a verbal agreement—export the transcript and store it yourself rather than leaving it live in the vendor's search index indefinitely.

4. Route exports to storage you control, not another vendor's cloud. This is the step most people skip. Exporting a transcript out of Limitless or Plaud and into a personal Google Drive folder just moves the exposure from one third party to another. Proton Drive gives you end-to-end encrypted storage under Swiss privacy law, with a folder structure you control—store exported transcripts there instead, and they're encrypted before they ever leave your device. The free tier covers casual use; Proton Unlimited adds enough space for years of exported audio and text at a reasonable monthly cost. The same "don't let the export just move the exposure" problem applies to any AI-generated file, not just transcripts—see what happens to the files your AI tools produce for the wider pattern.

5. For anything client-facing, business-sensitive, or legally privileged, skip vendor cloud storage entirely. If you're a consultant, attorney, or freelancer using an AI pendant to capture client meetings, the export-and-delete workflow above isn't strong enough on its own—you need storage where even the provider can't access the plaintext. Tresorit is zero-knowledge by architecture: exported meeting transcripts stored there are unreadable to Tresorit itself, which matters if a client's NDA or your professional privilege obligations require you to prove the data was never accessible to a third party. Tresorit also supports expiring, download-limited share links, which is the right way to send a transcript excerpt to a colleague without leaving a permanent copy in their inbox. If you're not sure whether Tresorit or Proton Drive fits your setup, Tresorit vs. Proton Drive compares the two directly.

A Quick Comparison

|---|---|---|---|

| Searchable by vendor | Yes, by design | No | No |

What About the People Who Didn't Choose to Wear One

The hardest part of this category isn't protecting your own data—it's the fact that everyone who talks to you while you're wearing one of these devices is having their voice captured without a device of their own to configure. That's not a technical problem you can solve with better storage. It's a disclosure norm you have to build for yourself.

A reasonable baseline: mute or remove the device in one-on-one conversations that are personal, sensitive, or professionally privileged, unless you've said out loud that you're recording. Treat the "always listening" mode as something you use in low-stakes contexts—your own voice memos, solo commutes, note-taking to yourself—rather than as a permanent default around other people.

What This Setup Gets You

Once the workflow above is in place:

The device still does what you bought it for—ambient capture, searchable notes, less manual note-taking
Training opt-in is off, so your conversations aren't feeding a model you didn't agree to train
Retention is bounded, so a breach five years from now doesn't expose a conversation you forgot existed
Anything you actually want to keep lives in storage where you—not the vendor—hold the keys
Client and privileged conversations get zero-knowledge storage, not vendor-default cloud

AI wearables are genuinely useful. The privacy problem isn't the microphone—it's the default assumption that everything the microphone hears should live forever on someone else's server. Fifteen minutes of settings changes and one storage decision fixes that.

Stay Current on Privacy Tools That Actually Matter

New wearables, retention policies, and legal precedents in this space are moving fast. Subscribe below for a monthly digest—no AI-generated filler, just signal from tools we've tested ourselves.

Subscribe to the PrivateAI newsletter

Last updated: 2026-07-12