A researcher found a way to signal to AI meeting bots that you do not want to be recorded, exploiting a gap in how tools like Otter.ai and Fireflies process consent and metadata. The hack does not block recording outright. It poisons the output, making transcripts unreliable or triggering opt-out logic baked into some bots by default.
This matters because the default in 2024 is surveillance. Every Zoom call, every standup, every one-on-one is a candidate for transcription and summary, often without explicit acknowledgment from all participants. The researcher's method is a countermeasure built from the same infrastructure these tools rely on, which is what makes it technically interesting and worth reading in full.
The real question the piece forces is not whether the hack works at scale. It is whether users have any meaningful recourse when recording is the assumed baseline. The answer, right now, is almost none.
[READ ORIGINAL →]