# How are AI-cloned voices used in election robocalls, and does anyone detect them in time? Source: https://global100.org/forum/q/how-are-ai-cloned-voices-used-in-election-robocalls-and-does-anyone/ Site: Global 100 Forum, category AI voice and audio Published: 2026-11-05. Updated: 2026-11-08. Replies: 8. ## Question by Kwame A., 2026-11-05 Election season brings back a worry that I cannot settle from reading. Voice cloning is now cheap and quick, and an automated call is the cheapest way to reach a lot of people. I remember the calls in January 2024 around the New Hampshire primary that imitated a well-known politician's voice. That was a few years ago, and I assume the problem has not gone away. I have three linked questions. First, how are cloned voices actually used in election calls: to impersonate a candidate, an official, someone local the voter would trust? Second, does anyone detect these calls before they reach people, or only after complaints come in? And third, what is the legal position in the US, since I have seen both "it is illegal" and "it is a gray area" said with equal confidence. I am trying to write something short for a student society about what a voter can actually do when the phone rings in the last week before an election. I do not want to give them a list of things to "watch for" that does not work. If the honest answer is that nobody can detect these in real time, then I would rather say that and give them something else to do. ## Reply 1 by Ingrid V., 2026-11-06 From the platform side: nothing detects these in real time for the ordinary voter, and I would say so plainly. The attacker needs a short clip, a script and a calling service. The defender needs to know a call is fake within seconds, on a compressed phone line, with no file to examine. That is a lopsided contest. What works at scale is not detection of the voice. It is the plumbing around the call: who is allowed to place calls, which carriers verify the numbers, how quickly complaints are aggregated, and how fast an operation can be shut off once it is spotted. ## Reply 2 by Jules, 2026-11-07 the quick fix is just to get a call-screening app that has an AI voice detector in it. they exist now and they flag cloned voices live while you are on the call. install one before the election and you are covered. ## Reply 3 by Sam Whitlock, 2026-11-07 Treat it as an authentication problem and not as a perception problem. A voter should not be asked to judge whether a voice is real. The voter should be asked to verify the claim through a channel the caller does not control. If a call says your polling place changed or the date moved, hang up and check the election office's own website or a number you already have. Same defence as the family emergency scam: do not trust the channel the message arrived on. ## Reply 4 by Hana Sato, 2026-11-07 @Jules I would not tell anyone that an app will cover them. Live voice detection on a compressed phone call is among the hardest cases, and public evaluations of voice detectors tend to show accuracy dropping on audio that is not clean studio quality. Even a good detector gives a probability and has false alarms, which on a call about where to vote could also scare people away from a real message. If the app flags the call, that is a hint to verify, not a conclusion. Our [thread on voice detector accuracy](/forum/q/how-accurate-are-ai-voice-detectors-in-real-conditions/) goes into it. ## Reply 5 by Rob M., 2026-11-07 @Sam Whitlock yes, exactly this. The family version of "check through a channel they do not control" was a code word and a call-back. Same shape. I think the student handout could say so in one line, since it works for both. ## Reply 6 by Rob M., 2026-11-08 My mother's call taught me this. Not one thing in that call sounded wrong to her. The trick was that it sounded right. So any advice that begins "listen for odd pauses or a robotic tone" is advice for a world that is mostly gone. What saved us was calling back on a number we already knew. ## Reply 7 by Tomas Reyes (staff), 2026-11-08 No, not in time for the voter, as far as public sources show. A live phone line gives no file and little time, and the defences that exist work around the call: carrier-level caller ID authentication, enforcement under US telephone law, and verifying claims through official sources. For voters, the reliable defence is procedural, not auditory. On the law, the FCC's February 2024 release, ["FCC Makes AI-Generated Voices in Robocalls Illegal"](https://docs.fcc.gov/public/attachments/DOC-400393A1.pdf), says the commission unanimously adopted a Declaratory Ruling recognising that calls made with AI-generated voices are "artificial" under the Telephone Consumer Protection Act. It was framed as a tool against voice-cloning scams, and the FCC's own statement lists extorting family members, imitating celebrities and misinforming voters among the harms. The release explains that this makes the use of an AI-generated voice in such calls itself illegal, giving state attorneys general another route to act, and that the TCPA lets the FCC fine robocallers and lets consumers sue. So "it is a gray area" is out of date for automated calls. What the release does not do is say anything about detecting such calls: it notes that the FCC had asked, in a separate inquiry, how AI might help recognise illegal robocalls before they reach consumers, which is a question, not a deployed system. On the network side, the FCC's [call authentication page](https://www.fcc.gov/call-authentication) describes SHAKEN/STIR, a framework in which the originating carrier "signs" the caller ID and other carriers validate it, so the receiving carrier can verify that a call is from the person making it. That targets spoofed numbers, not synthetic voices: a call can be properly signed and still use a cloned voice. It is a useful layer, and not a voice detector. For your student society: do not teach "listen for glitches". Teach "hang up and verify through the official election office". ## Reply 8 by Kwame A., 2026-11-08 @Tomas Reyes that is exactly the distinction I needed: the ruling covers the use of the voice, the authentication framework covers the number, and neither is a voice detector. The "hang up and verify" line will be the headline of the handout. I will quote the FCC wording rather than paraphrase it. --- Cite as: Global 100 Forum, "How are AI-cloned voices used in election robocalls, and does anyone detect them in time?", https://global100.org/forum/q/how-are-ai-cloned-voices-used-in-election-robocalls-and-does-anyone/, accessed 2026-11-10.