Threat Analysis: Deepfakes Targeting US Government Officials

Threat analysis: how deepfakes target US government officials

The FBI describes a long-running social engineering campaign in which attackers pose as senior US officials. On 15 May 2025, alert I-051525-PSA warned that since April 2025 malicious actors had been sending text messages and AI-generated voice messages, known as smishing and vishing, that claimed to come from senior officials. Many targets were current or former senior federal or state officials and their contacts. The stated goal was to build rapport and then gain access to personal accounts, often by sending a malicious link under the pretext of moving to another messaging platform.

On 19 December 2025, the FBI updated the warning in I-121925-PSA. It said the impersonation has been going on since at least 2023 and covers state, White House and Cabinet-level officials and members of Congress. Targets now include officials' family members and acquaintances. Once on an encrypted app, the impostors discuss policy, offer meetings with senior leaders or suggest board nominations, then ask for wire transfers, passport copies, two-factor authentication codes or access to contact lists, which they use to impersonate more people.

The Rubio voice impersonation case

The best-documented single incident involved Secretary of State Marco Rubio. In mid-June 2025, someone created a Signal account with the display name marco.rubio@state.gov, which is not a real address. According to a State Department cable reported by CBS News and others in July 2025, the impostor contacted at least five people outside the department: three foreign ministers, a US governor and a member of Congress. The actor left Signal voicemails for at least two of them and sent a text inviting another to talk on Signal.

The cable said the actor likely aimed to manipulate targets using AI-generated text and voice messages to gain access to information or accounts. The State Department said it was aware of the incident and investigating. No attribution was announced. The case shows the pattern the FBI describes: a plausible identity, a cloned voice that removes the usual sense that something is off, and a move to an encrypted app where official logging and security controls do not apply.

The attack chain, step by step

Breaking the campaign into stages shows where defenders can intervene. Each stage relies on a different weakness, and only one of them involves the deepfake itself.

The voice clone matters most at the trust-building stage. Senior officials give speeches, interviews and hearings, so there is ample public audio to train a convincing clone. A short voicemail in a familiar voice lowers suspicion enough for the target to reply or click. After that, the attack is classic account takeover and fraud.

Who is at risk beyond the officials themselves

The campaign aims at the network around an official as much as the official. The December 2025 alert explicitly includes family members and acquaintances, and the Rubio case reached foreign ministers, a governor and a member of Congress. Former officials, congressional staff, state government employees, foreign diplomats and executives who deal with government are all plausible targets, because each holds contacts and information the attacker can reuse.

Private companies face a related risk. The same techniques that impersonate a Cabinet member can impersonate a regulator, an agency contact or a government customer to push a supplier into sharing data or changing payment details. Government-facing businesses should brief staff that a call or voicemail from a known official is not proof of identity.

Controls that stop official impersonation

The FBI's guidance focuses on verification habits rather than spotting the fake. Its recommendations include independently finding a known phone number for the person and calling it, checking names, numbers and URLs for small changes, being skeptical of voice messages because cloning can sound nearly identical to the real person, and not clicking links until the sender is confirmed. It also advises enabling multi-factor authentication and never sharing a two-factor code with anyone, including over encrypted apps.

Organizations should turn those habits into procedure. Publish which channels senior leaders actually use, and state that they will never ask for codes, documents or money through personal messaging apps. Give staff a simple way to report a suspected impersonation to security and to IC3, and preserve the messages and voicemails before blocking the account.

• Call back on a number from an official directory, never one supplied in the message.
• Agree verification phrases within families of senior officials, as the FBI suggests.
• Keep government business on managed, logged platforms rather than personal apps.
• Report attempts to your security office, the local FBI field office or ic3.gov with full details.

Where detection fits for government security teams

Verification by callback does not depend on detection, and it should not wait for it. Detection becomes useful after a report, when security teams want to know whether a preserved voicemail or clip is synthetic, whether several incidents used the same cloned voice, and how to brief leadership and partners. A short, compressed voice note gives less evidence than a long recording, so findings should come with stated confidence.

NeuroDetect, Neural Defend's forensic media workspace, supports that investigation. Analysts upload authorized audio, video, images or documents into a case. Each analysis keeps its model and version context, detailed views drill down to frames and audio segments, and an LLM-assisted summary explains findings in plain language with links back to the evidence, while the analyst makes the final call. The downloadable forensic report records observations, uncertainty, provenance and analyst notes. Neural Defend supports private cloud, on-premises and air-gapped deployment for agencies that cannot send sensitive recordings to a shared cloud, and holds ISO/IEC 27001 and SOC 2 Type II certifications. For live phone calls, NeuroCall, currently in Beta, is Neural Defend's voice deepfake detection product.