For most of the telephone era, recognizing someone’s voice was useful evidence of identity.
It is weaker evidence now.
The FTC has warned that scammers can use short audio clips to clone a family member’s voice and place convincing emergency calls. The FBI has separately documented campaigns using AI-generated voice messages while impersonating senior U.S. officials.
See the FTC’s warning about AI voice-cloning emergency scams and the FBI’s 2025 impersonation alert.
Resemblance is no longer authentication
A convincing voice, face, or video can still be useful context.
It just cannot carry the entire burden of proof when the request matters.
If a familiar-looking executive suddenly requests a confidential transfer, or a relative’s voice asks for emergency cryptocurrency, the audiovisual resemblance is part of the claim being tested. It is not independent confirmation of that claim.
This is an awkward change because human recognition is fast and deeply practiced. We are used to treating “I heard her voice” or “I saw him on video” as the end of the identity question.
Synthetic media turns those observations into one signal among several.
Move verification outside the suspicious interaction
The FBI recommends independently confirming a claimed identity when something unexpected or consequential is requested. That can mean calling a known number, using an established organizational channel, or checking with another person who can verify the request.
For families, a previously agreed word or question can add another signal, but the broader principle matters more: do not let the suspicious call define the only path for confirming itself.
Deepfakes do not make identity unknowable.
They make passive recognition less sufficient.
A voice can sound right. A face can look right. A video call can feel immediate and personal.
The decision still deserves evidence that did not come from the same channel asking for the money, credentials, secrecy, or access.