Imagine sitting across from a friend at a coffee shop, talking about something personal. Maybe it’s a health concern, a work frustration, or a private hope. There’s a small device on your friend’s wrist, or clipped to their shirt, or snapped onto the back of their phone. It has a tiny light, but it’s easy to miss. That device is Amazon’s new Bee, and it’s listening. It’s turning your conversation into a summary and a to-do list that only your friend can see. Amazon says the Bee is a personal AI assistant, a helpful companion that remembers what you discussed and follows up on your behalf. The two new models, the $74.99 Bee Frontier and the $99.99 Bee Eclipse, promise to make life easier. But according to Jeff Williams, a veteran privacy and security engineer who spent years at Amazon Web Services, the Bee is asking the wrong person for permission. The person wearing the device is prompted to get consent. The people being recorded have no practical way to say no. That small distinction is at the heart of a bigger problem, one that turns a seemingly helpful gadget into a quiet threat to everyone around it.
Amazon unveiled the Bee hardware on a Thursday, the first new devices from the startup it acquired last year. The Bee Frontier is worn on the wrist or clipped to clothing; the Bee Eclipse attaches to the back of a phone. Both are designed to listen to conversations, extract the important bits, and turn them into summaries, reminders, and action items. For paying subscribers, the devices connect to Amazon.com and Alexa+, meaning the assistant can buy things, manage schedules, and integrate with Amazon’s broader ecosystem. It’s an ambitious vision, one that positions Amazon at the center of your daily life, not just through your phone or speaker, but through a device that travels with you everywhere. But that vision depends on recording conversations, and conversations are not one-sided. When the Bee records you, it also records the people you’re talking to. That’s the fundamental tension. Amazon’s design puts the wearer in control, with a prompt on their phone and a light on the device. But as Williams points out, everyone else in the conversation is left in the dark, literally and figuratively. Their words are captured, processed, and stored, yet they never agreed to any of it.
Williams, who worked as a senior principal security engineer at AWS and is now the chief information security officer at Sigma360, laid out his concerns in a LinkedIn article. He doesn’t question Amazon’s technical capabilities. He questions the basic ethics of the product’s consent model. Amazon says users are prompted to get permission from call participants each time they start a recording, and the devices show an indicator light while in use. But Williams notes that the prompt appears only on the wearer’s phone. There is nothing forcing the wearer to act on it. A user can tap “I’ll ask,” and then simply not ask. The light, meanwhile, is easy to hide under a sleeve, in a pocket, or behind the phone. On a video call, the wrist or phone is often out of frame. So the person on the other end of the call has no idea they’re being recorded. Williams calls this a “liability allocation,” a way for Amazon to shift the responsibility onto the wearer. In other words, Amazon can say it has a consent feature, and if the wearer ignores it, that’s the wearer’s fault. But the person being recorded never had a chance to consent in the first place. It’s like asking someone to hold up a sign that says “I promise not to lie” and then blaming them when they lie. The system looks good on paper, but it doesn’t work in practice.
This isn’t a hypothetical concern. Williams points to Meta’s Ray-Ban glasses as a cautionary tale. In September, the data protection authority in Hamburg, Germany, found that the glasses’ recording light was hard to see outdoors. The authority concluded that a tiny indicator light alone doesn’t establish informed consent under Europe’s General Data Protection Regulation, known as GDPR. If a device is recording, people need to know, not just in theory, but in reality. A light that’s barely visible in sunlight or hidden by a sleeve isn’t meaningful notice. In the United States, the legal landscape is patchwork. Roughly a dozen states, including Washington and California, generally require all parties to consent before a private conversation can be recorded. But in many other states, only one party needs to consent, which means the wearer could legally record without telling anyone. That creates a confusing and dangerous situation. The Bee might be legal in one state and illegal in another, but its design doesn’t change based on where it’s used. And even where it’s legal, legality isn’t the same as morality. People don’t expect a friend’s wristband to be capturing their words. They don’t expect a phone on the table to be listening while they speak. The burden is placed on the wearer to be honest, and honesty is not something you can enforce with a prompt.
Williams does give Amazon credit for one thing: Bee Private Compute. The system processes audio in an isolated, encrypted environment built on AWS Nitro, and then deletes it. Amazon says only the customer can access the resulting data, not Amazon or Bee. That’s a meaningful privacy feature, and it shows that Amazon is thinking about security. But Williams argues that this protection only shields the wearer. The people around the wearer have no such protection. Their words end up in transcripts and summaries that they can’t see, can’t access, and can’t delete. Amazon says those speakers aren’t identified by name unless they complete a voice training exercise with the device’s owner. But the words themselves are still there, tied to a moment, a voice, and a context. If someone says something sensitive, like “I’m worried about my medical test” or “I might lose my job,” those words become part of the wearer’s personal notes. There’s no mechanism for the other person to hear what was captured, correct a mistake, or ask for it to be removed. Williams also notes that Amazon has changed similar commitments in the past. Last year, the company ended the “Do Not Send Voice Recordings” option on some Echo devices, a move that undermined trust in its privacy promises. He advises companies to keep these devices out of client meetings and privileged discussions, because the transcripts end up in an employee’s personal account, where the employer can’t see or manage them. That’s a legal and ethical minefield, especially in fields like law, medicine, and finance, where confidentiality isn’t just a nice idea, it’s a legal obligation.
Williams has a proposed fix, and it’s surprisingly simple. He says Bee should leave other people’s voices out of the transcripts by default, unless they’ve enrolled their voices and agreed to be recorded. Before Amazon acquired the startup, Bee had indicated plans to record only the voices of people who had verbally consented, according to TechCrunch. That design would flip the default: instead of recording everything and hoping the wearer asks permission, the device would record only the person who opted in. Everyone else would be excluded, their words reduced to background noise or silence. This isn’t a radical idea. It’s a privacy-by-design approach, one that builds consent into the technology itself, rather than relying on a user’s goodwill. Williams says he knows first-hand that Amazon has the engineering talent to ship it. The company could use voice recognition, on-device processing, and other tools to ensure that the Bee only captures the voices it’s supposed to capture. Amazon didn’t comment directly on the critique, but it pointed to safeguards built into the devices. One is an automatic shutoff: the devices turn off if they don’t detect the owner’s voice for 5, 15, 30, or 60 minutes, depending on the user’s setting. Amazon says this reinforces that Bee is “a personal AI assistant, not a device designed to passively pick up audio throughout the day.” The company also says it educates customers about responsible use in the Bee app, and that the Eclipse will send push notifications encouraging users to let people around them know they’re using Bee and ask for permission. But these are soft solutions. The automatic shutoff only prevents recording when the owner is absent; it doesn’t stop the device from recording bystanders while the owner is present. The notifications still depend on the wearer to do the right thing. And the education in the app only reaches the person who bought the device, not the people who might be recorded. In the end, Bee puts the burden on the wearer to be honorable, and the people around the wearer are left hoping that the wearer is honorable. That’s not privacy. It’s trust. And trust is not something a device can enforce. The Bee is a clever, capable assistant, but until Amazon changes the design so that other people’s voices are excluded by default, it will remain a personal AI assistant that treats everyone else as background noise. For a feature called Bee, the sting is in the lack of respect for the people doing all the talking.
