Quick answer: Government and law-enforcement impersonation complaints doubled in 2025 per the FBI IC3. AI voice cloning has eliminated the audio tells consumers relied on. Spot a fake-agent call by listening for missing breath, a too-clean background, scripted urgency, and gift-card or wire demands. Verify by hanging up and dialing the agency's published number.
Your phone rings. The caller ID says "Internal Revenue Service." The voice on the line identifies as Agent Daniels, badge number 4471, with the Criminal Investigation Division. There is a warrant for your arrest over an unfiled return. You can settle the matter today, over the phone, by purchasing $2,800 in gift cards and reading the codes back. Or you can wait for the marshals, who will be at your address within the hour.
The agency is real. The badge number is fabricated. The voice is cloned. The call is the version of government-impersonation fraud that doubled in 2025, and the AI-enabled version is already the dominant variant.
This post walks through seven specific signs to listen for on a suspected fake-agent call, the 30-second flow you can run during the call, and the verification step that does not erode as model quality improves.
For the broader technical grounding on how synthetic voice gets generated and why the audio tells alone are no longer enough, see the voice cloning detection guide and the pillar guide on what a deepfake actually is.
Doubled in one year
Government and law-enforcement impersonation complaints reported to the FBI Internet Crime Complaint Center grew 100 percent in 2025 versus 2024. The growth is concentrated in calls that use AI voice cloning to impersonate IRS agents, Social Security Administration officials, local police, and federal marshals. The fraud-kit market has industrialized the workflow.
Source: FBI Internet Crime Complaint Center 2025 Annual Report; analysis covered in our FBI IC3 2025 breakdown.
Why Government-Impersonation Calls Are the Highest-Leverage AI Scam
Three structural reasons the fake-agent call is now the dominant impersonation vector.
The authority signal does the heavy lifting. A scammer claiming to be your cousin needs to construct rapport. A scammer claiming to be a federal agent inherits the rapport from the institution. The target's cognitive load shifts immediately from "is this real" to "what do I need to do." The voice clone only has to hold up long enough for the urgency framing to land.
Spoofed caller ID and cloned voice are now standard. Fraud kits documented in industry reporting bundle caller-ID spoofing, voice cloning trained on publicly available agency-spokesperson audio, scripted prompts, and target lists for low monthly subscription prices. The packaging is what changed in 2025. The technology has been available for years; the productization made it accessible to operators who do not have technical skill.
The voluntary AI label does not exist on the phone network. Phone calls carry no equivalent of the Instagram AI Info label or Meta's labeling regime. The detection happens entirely in your ear, in real time, while a person purporting to be a federal agent is engineering panic on the other end of the call. This is the structural gap that makes the call-side of fraud harder to defend against than the social-media side.
For the broader policy context, the EU AI Act Article 50 transparency rules cover AI-generated content but do not reach the phone network. The detection skill remains the floor.
Seven Signs to Listen For on a Fake-Agent Call
These signs help in 2026 against medium-quality clones. Strong current generators handle some of these better than others, so the absence of a single sign does not clear the call. Patterns matter more than individual cues.
1. The agency itself is wrong about its own procedures. Real IRS agents do not initiate contact by phone. The IRS sends mail first, then certified mail, then in-person visits with prior notice. Real SSA officials do not call to threaten suspension of your number; SSA numbers are not suspended. Real police do not call to demand payment to clear a warrant. The first sign that you are on a fake-agent call is that the call exists at all in a category where the agency never makes calls.
2. Payment method is gift cards, crypto, or wire to an unfamiliar account. No legitimate federal agency accepts payment via Apple gift cards, Google Play cards, Target cards, Bitcoin, or wire transfers to private accounts. The IRS accepts payment via IRS.gov, by check to the U.S. Treasury, or by enrolling in an installment plan. SSA does not accept payment to resolve account issues because there are no payments. A demand for any of the gift-card or crypto rails is the second-strongest tell on the call.
3. Breath is missing or fake. The same audio signature documented in our voice cloning detection guide applies to fake-agent calls. Real callers breathe between phrases, often audibly through the phone microphone. Cloned voices skip breath or insert flat silences that do not match the cadence of natural breathing. A federal agent with no audible breath in 30 seconds is a strong indicator the audio is generated, though high-quality clones increasingly fake breath as well; combine with the other signs.
4. The acoustic background is wrong for the claimed location. A real federal office is not silent. There are other agents talking, phones ringing, HVAC running, footsteps. A cloned voice claiming to call from a federal building in a synthetic acoustic vacuum is the second audio tell. The absence of background sound is harder to fake than the voice itself.
5. The script does not bend under unscripted questions. Ask for the agent's office address. Ask which IRS service center handles your taxpayer ID. Ask the spelling of the supervisor's name. Ask for the case number and what regional office it originated in. Real agents will answer these directly; scripted scam calls hit silence, change subject, or escalate the threat to keep you off the question. The latency on the response is the audio signature of either a clone struggling to generate plausible output or a human operator searching the script.
6. Urgency is artificial and reset-proof. The call constructs a deadline (marshals arriving within the hour, warrant active within 30 minutes, refund window closing today) that does not match how any federal process works. Federal enforcement does not move on the timescale of a phone call. When you propose hanging up and calling back, the urgency intensifies; the call cannot survive a 10-minute pause because the framing depends on you being inside the panic window.
7. They forbid you from hanging up. The single most reliable test. Tell the caller you need to call them back through the agency's published number. A real federal employee will accept this without resistance. A scammer will manufacture reasons you cannot hang up: the warrant becomes active the moment you disconnect, you will be obstructing justice, you will lose access to your settlement window, you must stay on the line for legal protection. The resistance is the tell. Hang up anyway.
Think you found an AI video?
Paste the URL and let the Ledger community verify it. Free.
The 30-Second Listening Flow for a Fake-Agent Call
A scannable workflow you can run on any suspicious government-agency call. The clock is the structure your brain runs while the caller speaks.
- 0:00–0:05: First 5 seconds. Identify which agency the caller claims to represent. Confirm whether that agency contacts the public by phone for the matter at hand (the IRS, SSA, and most federal agencies do not initiate by phone for enforcement).
- 0:05–0:10: Listen for breath. Listen for ambient sound behind the voice. A federal office is not silent; a cloned voice is.
- 0:10–0:20: Ask one off-script question. The supervisor's full name and spelling, the regional office address, the case number with the correct internal format. Time the response. A real agent answers in under 2 seconds; a clone or script search stretches to 3 to 5 seconds.
- 0:20–0:25: Listen for the payment method. Gift cards, crypto, or wire transfer to a private account are the strongest single tell. No legitimate agency accepts these.
- 0:25–0:30: Tell them you will call back from the number published on the agency's official website. Listen for resistance. Hang up regardless of their response.
If two or more signs fail, treat the call as a scam and hang up. The flow is short enough to run in your head while you continue speaking on the line. The pressure to comply inside the call is the operator's only leverage.
What to Do When You Suspect a Fake-Agent Call
Three steps in order.
Hang up. Do not finish the call. No information transfer (no Social Security Number, no taxpayer ID, no date of birth, no bank routing number), no payment commitments, no agreement to stay on the line. The social cost of hanging up on a real federal employee is recoverable in 60 seconds by calling the agency back. The financial loss from a real scam is not. Treat the hang-up as the default response, not the escalation.
Call back through the agency's official published number. Not the number that called you, which is spoofed. Not a number a stranger gave you on the call. For the IRS, the published number is 800-829-1040 (individual taxpayers) listed at IRS.gov. For SSA, the published number is 800-772-1213 listed at SSA.gov. For local police, look up the non-emergency line on the police department's official website. If the original caller resists this when you propose it, you have the confirmation you need.
Report the call. Report to the FTC at reportfraud.ftc.gov. Report to the FBI IC3 at ic3.gov. For tax-specific impersonation, report to the Treasury Inspector General for Tax Administration at tigta.gov. For elderly victims, AARP runs a fraud hotline at 877-908-3360. Reports feed the pattern data that drives the federal enforcement priorities documented in the IC3 2025 report breakdown; underreporting is the single biggest gap in the policy response.
If a relative was targeted and the caller used a cloned voice of a family member to layer the scam (a "grandchild in jail" version that pairs with the fake-agent call), the family-protection playbook covers the safe-word protocol and the coaching script for older relatives.
What Ledger Does Differently
The fake-agent call operates outside the platforms Ledger checks today. The phone network has no AI label, no community flags, no detection layer. The defense on a call is yours: the listening flow, the callback, the hang-up.
What Ledger covers, and what compounds with this guide, is the visual side of the same operator pattern. The scammer who runs a fake-IRS call often runs a fake-agency video alongside it (a "settlement portal" video on TikTok, an Instagram Reel of a "compliance officer" walking through fake instructions, a YouTube ad for a fraudulent debt-resolution service that funnels victims to the call). Paste any URL into the free AI video detector to see whether the Ledger community has already flagged the account behind it.
The call you got on the phone often has a video counterpart on social media. If you were directed to verify your case at a URL, watch a "settlement walkthrough" before paying, or confirm your identity by clicking through a video link, that video is the searchable side of the operator. Check it.
If you want to help build the community record that catches operator patterns across video surfaces, join the iOS or Android waitlist and be among the first to flag accounts when the apps ship.
[APP-DOWNLOAD]
Related Posts
- How to Tell If a Voice Is AI-Generated: 7 Signs to Listen For Before You Trust Audio: the audio-detection sibling that covers the broader voice-cloning signal set, including the prosody and sibilance tells that carry into fake-agent calls
- AI Voice Cloning Scams Hit 1 in 10 Americans. Here Is How to Protect Your Family.: the family-protection sibling covering the safe-word protocol and the coaching script for relatives in the target window for fake-agent calls
- FBI IC3 2025 Report: AI-Enabled Fraud Grew 1,210% in One Year. Here Is What the Numbers Show.: the primary-source federal data behind the doubling of government-impersonation complaints in 2025
- What Is a Deepfake? A Plain-English Guide for Social Media Users: the technical foundation explaining why source-audio requirements have collapsed and why voice cloning is now within reach for unskilled operators
