Sat 21 Nov 2026 / 15:46 ET
Kernel
Long Reads 3 min read

EFF says ambient AI devices leave bystanders without a privacy choice

EFF argues that smart glasses and AI transcription tools put notice and consent burdens on people who do not own them.

Theo Lindgren

By Theo Lindgren / Columnist

EFF’s ambient AI privacy argument is aimed at a growing class of devices that can observe, record, transcribe or summarize the world around their owners. In a Nov. 13 post, the digital-rights group said the practical privacy problem is no longer limited to what a device owner agrees to share. It also reaches the people within a camera’s view or a microphone’s range.

EFF cited Meta and Google smart glasses, Apple Watch audio transcription or summarization features, and Amazon’s Bee recording and transcription devices. The organization’s case is an advocacy argument, not an independent finding that every product works the same way or that every company is deliberately remaking privacy rules.

Its central complaint is more concrete than the broad rhetoric around “AI privacy.” Device makers commonly offer controls, limits on company access, and other safeguards to their customers. Those settings may matter. But EFF argues they do not give a nearby colleague, friend, stranger or family member a comparable way to decline capture.

How do ambient AI devices affect bystander privacy?

EFF breaks the issue into a problem of notice and refusal. A person who does not own the device may need to recognize an unfamiliar pair of glasses, watch, pendant or other wearable; know whether it is recording or analyzing audio or video; understand how it indicates that activity; and determine what service receives the resulting data. That is a tall order for consenting to a conversation at a table, on a street, or at work.

The group distinguishes this from someone visibly taking out a phone or camera to record. Those older tools can also document conversations, but EFF says they generally involve an overt action that people nearby can recognize. Ambient products are designed to operate in the background, which makes disclosure by the user the main safeguard for people around them.

Apple points to technical barriers between the company and data generated by its Audio Intelligence feature, according to EFF. Meta and Google describe privacy controls for their glasses, and Amazon says Bee users are in control and that no audio is stored. EFF does not argue that these claims are irrelevant. Its objection is that such protections are chiefly framed around the product owner rather than the non-user who is captured.

What happens to transcripts and summaries?

EFF also raises a downstream question: even where raw audio is discarded, a transcript, summary or other derived record may remain. The organization says such records could become targets for law-enforcement access. The available material identifies no particular subpoena, case, retention schedule or product-by-product policy, so that point remains a prospective concern rather than a documented outcome.

Google’s responsible-use terms tell users not to record people without permission or in sensitive places, EFF noted. The group argues that terms directed at users do not solve the practical consent problem for bystanders who may not know a device is active.

The broader concern is social rather than technical. EFF predicts that routine, inconspicuous documentation could make private conversations less candid and weaken the norm against preserving every interaction. That prediction has not been measured in the material reviewed here. Still, it identifies the pressure point companies’ privacy settings tend to skip: one customer’s choice can create a record about somebody else.

The dispute over privacy messaging has older roots. A 2019 review by the International Association of Privacy Professionals found large consumer technology companies regularly used public statements and marketing materials to explain their privacy approaches. The review explicitly evaluated that communication, not the companies’ underlying privacy practices.

This story draws on original reporting from Techdirt.

More Long Reads/

view all ↗