Sub-Audible Signal Standard for Audio Capture and Transcription Opt-Out (dontrecord.me)

Security researchers George Danforth and Jonathan Mortensen propose "DO NOT RECORD ME" (DNRM), a standardized sub‑audible digital signal that individuals can inject into audio streams to assert an opt‑out from AI audio capture and transcription. Framed as an audio equivalent of the HTTP Do Not Track header, DNRM is intended to be detected by compliant transcription systems so the originating speaker’s audio is excluded from automated transcription. The proposal responds to the rapid spread of low‑cost, AI‑driven transcription tools and the lack of a technical way for nonconsenting speakers to assert privacy preferences. Technically, DNRM is an infrasonic composite of three sine waves at 2 Hz, 3 Hz, and 5 Hz with an amplitude constant of 0.0025 (kept below −50 dB) so it remains effectively inaudible on typical consumer devices yet is preserved by common 16 kHz digital audio pipelines (infrasonic content is oversampled there, while ultrasonic signals would be undersampled). Detection is described as straightforward signal processing (e.g., narrowband/spectral detection or matched filtering). Key implications: the approach is lightweight and interoperable but depends entirely on vendor compliance, faces real‑world robustness issues (microphone/codec low‑frequency response, ambient noise, false positives/negatives), and raises legal and adversarial considerations that will shape its practical adoption.