Editorial workflow question. were running phone interviews with sources and need to verify source identity isnt being spoofed by AI voice cloning. high-stakes interviews only (whistleblower, expert testimony, witness accounts). current method: video confirmation at the start of the call. but some sources cant or wont use video. whats your workflow for audio-only source verification in 2026
Several methods working in practice for audio-only verification: 1) Two-factor identity confirmation. Source confirms identity via a second channel before the call (signal message from a verified number, or a verification code through a known email). 2) Live conversational verification. Ask 2-3 questions during the call that only the real source would know, ideally questions that reference earlier interactions or detailed personal context that hasn’t been published. 3) Voice biometric matching if you have prior recordings of the source. Pindrop and similar services can do this against historical audio. 4) Background context check. Real callers slip in unconscious environmental cues (specific room noise, regional accent variations) that AI voice cloning often misses. Trained interviewers learn to listen for these. Layer at least 2 of these for high-stakes interviews. Single-method verification is increasingly unreliable.
The live conversational verification approach is what most academic interview protocols are moving to. References to context that hasnt been published is the strongest signal because AI cant fabricate context it hasnt seen. Works across languages too, unlike voice biometrics which have multilingual bias issues we discussed in another thread.
Adopted method 1 (two-factor) across our podcast network earlier this year. Source sends a signal message confirming theyre about to call from X number at Y time. Takes 30 seconds, ends entire class of impersonation attempts.
@JaxOnFire the layered framework is what we needed. were going to standardize on two-factor + conversational verification as the baseline. voice biometrics for cases where we have historical audio. writing this up as internal SOP this week. thank you.