If you're choosing between Amazon's new Bee wearables and the Friend pendant, the most consequential difference isn't the form factor or the price—it's where your conversations go. Amazon says its new Bee Frontier and Bee Eclipse models process speech on the device itself under a "Private Compute" architecture and store no audio at all. Friend's own privacy policy, by contrast, names OpenAI and xAI as processors of your device audio and conversation context. That single distinction should drive the decision for anyone whose recordings touch confidential settings.
Amazon Bee: On-Device Private Compute for Data Minimization
Amazon introduced Bee Frontier and Bee Eclipse on October 15, 2026, and the company frames privacy as a design principle rather than an afterthought. Both devices are built on what Amazon calls "Bee Private Compute," described in its announcement as "a security architecture that processes your conversations inside a secure and isolated processing environment where no one, including Amazon, can access your data."
The practical claim is blunt: "No audio is ever stored." Independent analysis of the device's policies aligns with that position, reporting that Bee does not train on user data and that audio is never stored. One caveat from that same secondary analysis: the Bee Pro tier links to Alexa+ and Amazon.com, which is worth separating from the core device's audio handling when you evaluate the full ecosystem.
Friend Wearable: Cloud-Based Processing and Third-Party AI
Friend's privacy policy takes the opposite approach. It names OpenAI as a processor that "processes Device audio and friend context in order to power real-time spoken conversations," and may also process chat content to generate the friend's text replies. X.AI is named as another processor of device audio and friend context. The policy adds that which provider handles a particular conversation depends on the feature in use and on Friend's operational routing, and that this "may change over time."
That means your audio leaves the device by design. Secondary analysis of the policy also flags a contradiction: Friend's marketing suggests nothing is stored, while the policy itself contemplates retention of over five years. The same analysis reports that the policy names Google Cloud AI and service-provider sharing, and that it may collect voice patterns and potentially biometric, health, and geolocation data. Training on user data is described as anonymized with an opt-out in settings, per that secondary reporting.
Comparing Privacy Architectures: Bee's Local vs. Friend's Cloud
Strip away the marketing and the two devices represent cleanly opposed data-handling models. Bee's architecture concentrates processing inside the device in an isolated environment that even Amazon says it cannot access, minimizing what exists to leak. Friend's architecture distributes your audio across at least two external AI providers whose routing can shift without notice, which multiplies the parties you must trust and makes retention terms the decisive question rather than the architecture itself.
Neither model is automatically "secure" or "insecure"—they simply place control in different hands. With Bee, the trust question is whether you believe Amazon's on-device claims. With Friend, the trust question is whether OpenAI's, xAI's, and Friend's retention and training practices match your tolerance, and whether the marketing-versus-policy contradiction noted by outside reviewers matters to you.
AI Wearable Privacy Comparison: Bee vs. Friend
The table below juxtaposes Amazon's official claims against Friend's policy language and secondary policy analysis, isolating the core differences in data control.
| Dimension | Amazon Bee (Frontier / Eclipse) | Friend Wearable |
|---|---|---|
| Stated data processors | Conversations processed locally on the device, per Amazon's announcement | OpenAI processes device audio and friend context; xAI also processes device audio and friend context, per Friend's privacy policy |
| Audio storage policy | "No audio is ever stored," per Amazon; secondary analysis agrees audio is never stored | No explicit raw-audio retention statement found in policy analysis; marketing says none stored while the policy contemplates over five years |
| Privacy architecture claim | "Bee Private Compute": a secure, isolated processing environment where no one, including Amazon, can access your data | Cloud-based processing routed to external AI providers; routing depends on the feature and may change over time |
| User data for model training | Does not train on user data, per secondary policy analysis | Trains on anonymized user data with an opt-out in settings, per secondary policy analysis |
Read down the columns and the pattern is consistent: Bee minimizes what leaves your possession, while Friend externalizes processing and asks you to accept third-party retention terms that its own marketing language doesn't fully mirror.
Practical Implications for User Scenarios
Where you wear the device matters as much as which one you buy. For personal, low-stakes use—remembering a grocery idea, capturing a stray thought—Friend's cloud model is a reasonable trade, especially given its companion-style conversational features. The ambient-capture use case is different: a device that listens continuously in a home, a client meeting, or around employees changes the calculus, because the people nearby never consented to anything.
Secondary analysis of Friend's policy makes the same point from the other direction: the best use is personal recall where everyone around the wearer understands the device, and the worst is silent capture where consent, confidentiality, or professional duties are in play. For those higher-stakes settings, a device whose audio never leaves the hardware—assuming Amazon's claims hold—removes an entire category of third-party exposure. Note that this is a comparison of stated policies and architectures, not a hands-on test of how either device actually behaves in the field.
Making an Informed Choice: Aligning Wearable AI with Your Privacy Needs
Users should choose an AI wearable based on a careful assessment of its data processing architecture and their personal or professional confidentiality requirements. If your recordings will touch client conversations, workplace discussions, or anything you'd hesitate to email a stranger, Bee's stated on-device model gives you fewer parties to trust. If your use is personal and you value Friend's always-on companionship, the cloud model may be acceptable—provided you read the policy's retention and training terms yourself and use the opt-out if you object.
Before buying either device, check one measurable indicator: whether the official privacy policy explicitly states on-device processing for sensitive data and a clear "no audio storage" policy, or conversely transparently lists all third-party processors and their data retention and training practices. Friend's policy already names OpenAI and xAI; whether the company reconciles its marketing's "nothing stored" claim with the policy's five-year retention language is the specific clarification worth watching for before you commit.
