We Can’t Just Change the Definition of ‘Recording’
Tech companies with new AI hardware want you to agree that always-listening, often-recapping devices don’t count as recording.
By Victoria Song
As AI hardware proliferates in 2026, a troubling trend has emerged: companies are redefining what constitutes a “recording.” From pendant-like AI companions to smart glasses that continuously capture ambient audio, these devices are marketed as “memory aids” or “personal assistants” — not recording devices. But that semantic shift has significant legal and ethical implications.
The core of the issue is consent. Traditional recording laws — both federal and state — hinge on whether a conversation is being captured. If a device is always listening and storing snippets of speech, it is recording, regardless of whether the user thinks of it that way. By reframing the act as “recapping” or “remembering,” companies sidestep the need for clear disclosure and two-party consent in jurisdictions that require it.
In 2026, we’ve seen several high-profile incidents where AI wearables inadvertently captured sensitive conversations in workplaces, medical settings, and even private homes. The defenses have been predictably semantic: “The device only processes audio locally and discards it,” or “It doesn’t store full recordings, just summaries.” But summaries are still derived from recorded audio, and the initial capture is still a recording. The distinction is a distinction without a difference.
Worse, many of these devices are designed to be always-on by default. Users must opt out of continuous listening, rather than opt in. That inverts the privacy calculus that has governed recording devices for decades. A tape recorder only records when you press a button; an AI pendant might be recording from the moment you put it on.
Regulators are starting to take notice. In early 2026, the FTC opened inquiries into several AI hardware makers over deceptive marketing claims about recording capabilities. State attorneys general in California and Illinois have also signaled that existing wiretap laws may apply to always-listening AI devices, regardless of how companies label them.
But legislation lags behind the technology. Companies know this, and they are actively lobbying for definitions that exclude their products from recording statutes. They argue that because the audio is processed by AI and not stored as a traditional file, it shouldn’t be considered a recording. That argument is both technically and legally dubious.
The solution is not to abandon nuance. There are legitimate differences between a device that uploads full audio to the cloud and one that processes everything on-device and stores only transcriptions. But those differences should inform how we regulate, not whether we regulate. The baseline requirement should be transparency: if a device captures audio that could include other people’s voices, it is recording. Full stop.
We can’t let tech companies redefine fundamental terms to suit their business models. A recording is a recording, whether it’s on tape, on a hard drive, or in a neural network. The definition shouldn’t change just because the hardware got smaller and the marketing got smarter.
via The Verge AI