Marco Rubio Deepfake 2025: AI Spyware Targets Officials

Last reviewed: October 9, 2026. What changed: The FBI's Internet Crime Complaint Center issued a second public service announcement on September 17, 2026 (I-091726-PSA, "Scammers Impersonating Law Enforcement and Government Officials in Fraud Schemes") reporting nearly 61,000 complaints and over $1.6 billion in losses from law-enforcement and government-official impersonation between January 2025 and July 2026, and explicitly noting that scammers are using AI to appear as officials during video calls. The June 18, 2026 Blackburn-Coons bipartisan deepfake bill passed the Senate Judiciary Committee, the first federal legislative move of 2026 on political deepfakes; the same window saw the New York Times, citing Vanderbilt's follow-up to the GoLaxy documents, report on June 1, 2026 that China is now using AI to predict which of its own citizens could pose a political risk. The October 2, 2026 Axios story on AI campaign deepfakes drawing legal threats (Rep. Van Orden cease-and-desist) is the first 2026 US election-cycle instance in which a member of Congress has been on the receiving end of a deepfake-related legal demand. The September 8, 2026 findings (Rubio-class campaign dates to 2023; IC3 deepfake-portal re-victimization) remain intact.

Update (September 8, 2026): Two FBI actions since the July 2, 2026 update confirm that the deepfake-espionage playbook described in this article is still scaling up rather than tapering off. First, on December 19, 2025 the FBI revised the public service announcement it had originally issued on May 15, 2025, extending the campaign's timeline back to 2023 and adding detail on tactics (SMS outreach, escalation to Signal/WhatsApp/Telegram, requests for passport photos and overseas fund transfers, exploitation of stolen contact lists to expand targeting of White House, Cabinet, state-government, and congressional figures).^[12] The 2023 origin point means the original Rubio impersonation campaign is the surface of an operation at least two years older than the May 2025 alert window implied. Second, on July 20, 2026 the FBI's Internet Crime Complaint Center issued PSA I-072026-PSA warning that scammers are now using AI-generated deepfake videos of a senior FBI leader, hosted on a social media platform, to drive victims to spoofed IC3 lookalike sites that harvest names, phone numbers, emails, and loss data.^[13] The IC3 PSA also describes a parallel pattern in which prior fraud victims are re-targeted through Facebook Messenger and Telegram accounts posing as FBI agents, with the deepfake video serving as the credibility anchor for the spoofed portal. Together the two updates demonstrate that the threat model has matured in two directions at once: state-shaped precision espionage against political targets (the original Rubio scenario), and commodity-scale fraud that uses the same AI video tools against the very law-enforcement apparatus that investigates them.

Update (October 9, 2026): Four developments in the September 8 - October 9 window harden the threat model in two new directions: an official FBI tally of the impersonation economy, a federal legislative response that is now past its committee, a Vanderbilt follow-up showing the same AI-profiling infrastructure is now being turned inward against Chinese citizens, and the first 2026 US election-cycle instance of a member of Congress being on the receiving end of a deepfake-related legal demand. First, on September 17, 2026 the FBI's Internet Crime Complaint Center issued a second 2026 PSA, I-091726-PSA "Scammers Impersonating Law Enforcement and Government Officials in Fraud Schemes," reporting that between January 2025 and July 2026 the IC3 received nearly 61,000 complaints of law-enforcement and government-official impersonation with losses of more than $1.6 billion.^[14] The PSA explicitly names AI-generated video calls as one of the tradecraft vectors, lists jury-duty, license-expiration, medical-license, and foreign-targeted schemes as sub-categories with their own loss totals (jury-duty scams: 6,833 complaints, ~$36M; medical-license scams: 3,322 complaints, ~$37M; foreign-targeted scams: 1,809 complaints, ~$140M), and updates the March 2022 PSA the IC3 had originally issued on the same problem. The September 17 PSA is the FBI's first published number tying AI-enabled impersonation to a specific complaint volume, and the 60,940-complaint / $1.6B-loss tally is the first time the agency has put a public price tag on the impersonation economy the Rubio-class incidents live inside.

Second, on June 18, 2026 the Senate Judiciary Committee approved the Blackburn-Coons bipartisan bill to protect individuals and creators from deepfakes, the first federal legislative move of 2026 on the political-deepfake problem this article flagged in the fragmented-policy-response section.^[15] The bill's committee passage is the first time the 119th Congress has cleared a deepfake-specific measure through a Senate committee, and the committee action is now the floor under any further federal movement on political deepfakes in 2026. The bill does not yet resolve the fragmented-policy-response structure the article described, but it is the first procedural step toward the federal floor the article argued was still missing.

Third, the New York Times reported on June 1, 2026, citing Vanderbilt's follow-up to the GoLaxy documents, that China is now using AI to predict which of its own citizens could pose a political risk.^[16] The Vanderbilt follow-up is significant because the original GoLaxy research this article cites was about Chinese AI profiling of US targets; the June 1 NYT piece reports the same AI-profiling infrastructure is now being used to identify and target Chinese citizens. The same tradecraft the article describes as a foreign adversary capability is now being turned against the adversary's own population, which materially expands the dataset of "foreign adversary" profiling the article flagged and removes any clean distinction between the AI-profiling export and the AI-profiling domestic use case.

Fourth, on October 2, 2026 Axios reported that Rep. Derrick Van Orden (R-Wis.) received a cease-and-desist letter over an AI deepfake he posted falsely depicting his Democratic opponent, the first 2026 US election-cycle instance in which a sitting member of Congress has been on the receiving end of a deepfake-related legal demand.^[17] The same Axios coverage notes that as of June 2026, 31 states have election deepfake laws on the books, the first time the state-level count has cleared the 30-state threshold. The Van Orden case is the bridge between the Blackburn-Coons federal action and the state-level patchwork: the cease-and-desist will be argued under a state statute rather than federal law, and the state-count threshold means the federal bill this article flagged in the fragmented-policy-response section is now operating against a backdrop of more than 30 state regimes rather than the 19-20 states that were on the books when the original article was drafted. No new named impersonation of a sitting U.S. secretary on the Rubio template has surfaced since the December 2025 FBI revision; the underlying trend ("deepfake as a precision instrument of state influence, as a fraud vector against the agencies that investigate it, and now as a domestic political instrument against members of Congress") remains intact.

TL;DR: AI deepfakes have evolved from mass disinformation tools to precision espionage weapons, confirmed by the impersonation of U.S. Secretary of State Marco Rubio to deceive foreign officials. Foreign adversaries are now combining deepfake technology with detailed psychological profiling of over 2,000 American thought leaders and 117 members of Congress, enabling surgical manipulation of high-level diplomatic and policy decisions.

The Shift to Precision Espionage

Deepfakes (highly realistic synthetic media including videos, audio clips, and images) have moved beyond generalized misinformation campaigns to become core tools for state-level espionage.^[1] These AI and machine learning technologies can precisely mimic a person's voice, facial expressions, and mannerisms, blurring the line between authentic and fabricated reality.^[1] The scale of this threat is exponential: the number of deepfake videos shared online is expected to reach 8 million in 2025, a massive increase from 500,000 just two years prior.^[1]

This technological leap has changed the objective of digital deception. Unlike traditional intelligence gathering, which seeks specific secrets, the goal of modern AI influence operations is often to damage the fabric of trust itself, deceiving targets and sowing doubt within decision-making institutions.^[2]

High-Stakes Impersonation of U.S. Officials

The most alarming recent incident involved the confirmed use of AI deepfake audio recordings to impersonate U.S. Secretary of State Marco Rubio.[3, 2] In a series of targeted communications to senior government officials and three foreign ministers, an individual posing as the Secretary used AI tools to convincingly mimic his voice and writing style.^[3]

The intent was highly specific: to gain access to information and accounts, and potentially manipulate high-level state-to-state relations and policy decisions.[2, 3] This incident confirmed that generative AI is now a prioritized technological deception capability for adversarial states, targeting some of the most visible and critical posts in the U.S. government.^[3]

This pattern of targeting is not isolated. Last year, Senator Ben Cardin (D-MD) was deceived by a deepfake of a foreign official, an incident suspected to be tied to the Russian government seeking sensitive information about Ukrainian weaponry. Such cases confirm that elected officials now serve as both the direct target ("bullseye") for disinformation and the source material ("bait") for deepfakes used to deceive others.^[4]

Psychological Warfare: The GoLaxy Revelation

The effectiveness of these high-stakes deepfake attacks is enabled by the marriage of generative AI with precision psychological profiling. Reports from researchers at Vanderbilt University's Institute of National Security revealed that a Chinese company named **GoLaxy** has employed AI to construct detailed psychological dossiers on thousands of key American figures.[4, 5]

The scope of this data collection is a major national security concern, with profiles compiled on over 2,000 American thought leaders and, critically, at least 117 members of Congress.[4, 5] By analyzing this mass data, foreign adversaries can tailor propaganda and deepfake communications with surgical precision for maximum manipulative effect, ensuring the message and delivery method are optimized for the individual target.[4, 5] This capability represents a "fundamental shift" in how foreign influence can be projected into American civic life, moving from generalized propaganda aimed at the crowd to individualized, targeted deception.^[5]

Compounded Risks: Corporate Fraud and Obsolete Defenses

The threat extends far beyond the diplomatic and political spheres, posing immediate and severe risks to the private sector:

Deepfakes are now routinely integrated into sophisticated social engineering attacks. Criminal actors can generate the voice of a CEO to urgently request a fraudulent financial transfer, leading to severe consequences.^[1] Organizations are facing substantial financial losses, including reports of banks losing an average of $600,000 per voice deepfake incident, and a Hong Kong cryptocurrency scam facilitated by a deepfake voice that cost $18.5 million.^[6] Top organizational risks include financial losses, data breaches from bypassing security safeguards, and severe reputational damage.^[1]

The Obsolescence of Digital Defenses

One of the most concerning challenges is the speed at which deepfake generation tools are evolving. Detection tools developed even a year ago (which relied on identifying anomalies like unrealistic blinking rates or a lack of a heartbeat in subjects) are already obsolete.^[7] Advances in generative models and systematized approaches, where multiple AI models are used to correct the mistakes of others, are rapidly eliminating these visual and audio cues.^[7] This makes real-time, algorithmic liveness detection a formidable, ever-moving target for those tasked with securing the homeland.^[7]

Fragmented Policy Response

The U.S. federal regulatory response to AI deception remains fragmented. While some U.S. states have taken action (Montana, for instance, has enacted legislation regulating the use of deepfakes in election communications and criminalizing fabricated sexually explicit images) a comprehensive federal strategy to counteract political deepfakes is still developing.[8, 9] Congress and the Supreme Court face the delicate challenge of creating effective security standards to identify and counteract manipulated content while balancing these necessary measures against fundamental First Amendment rights.^[9]

Actionable Recommendations for Digital Trust and Security

Implement Call-Back Verification for Executives

Mandate strict, layered protocols for all high-value or urgent communications, especially financial transactions requested by executive personnel. A simple call-back or secondary, non-digital channel verification is essential, as AI can mimic voice and text with high fidelity.^[1]

Adopt Content Provenance Standards

Support and integrate emerging industry standards for content provenance, such as those promoted by the Content Authenticity Initiative (CAI) and C2PA. These standards aim to provide cryptographic "Content Credentials" that track the origin and history of digital media, allowing users and systems to verify content has not been surreptitiously manipulated.^[10]

Assume Zero-Trust for Digital Communication

Recognize that deepfakes have neutralized traditional trust signals like a known voice or a familiar face. Employees and security systems must adopt a zero-trust model for high-stakes digital communication, requiring technical or procedural authentication beyond simple recognition.^[2]

Limit Digital Footprint (Personal Security)

Be acutely aware that personal data (including social media activity, public statements, and professional biographies) serves as the "raw material" for AI profiling operations like GoLaxy. Limit the volume and intimacy of data shared online to reduce the efficiency of adversarial systems attempting to build your psychological profile.[5, 4]

References

Lawmakers are ‘bullseye and bait’ in AI-driven deepfake campaigns. Foundation for Defense of Democracies (FDD).
AI voice deepfake of US Secretary of State triggers global security alert. Biometric Update.
Deepfake: The War on Trust. The Cipher Brief.
The rise of deepfakes in cybercrime. Arthur J. Gallagher (AJG).
2025 Deepfake Detection Market Report & Buyer's Guide. Biometric Update (Market Report).
Vanderbilt Institute of National Security Releases Archive of GoLaxy AI Propaganda Research. Vanderbilt Institute of National Security.
Feature Article: Striking at the Heart of Adversarial AI. Department of Homeland Security (DHS).
Impacts of Adversarial Generative AI on Homeland Security. Department of Homeland Security (DHS).
Artificial Intelligence 2025 Legislation. National Conference of State Legislatures (NCSL).
Content Authenticity Initiative (CAI) Blog. Content Authenticity Initiative.
Deepfake. Wikipedia. Documents the October 2025 Catherine Connolly deepfake incident (RTÉ-spoof AI clip falsely claiming withdrawal from the Irish presidential race) and the June 2025 Sara Duterte deepfake incident (Philippine vice-presidential candidate).
FBI says 'ongoing' deepfake impersonation of U.S. officials dates back to 2023. CyberScoop. December 19, 2025. Reports the FBI's revised PSA extending the senior-official impersonation campaign timeline back to 2023 and detailing AI voice cloning via SMS, Signal, WhatsApp, and Telegram.
FBI Warns of Scammers Impersonating the IC3. FBI Internet Crime Complaint Center. Public Service Announcement I-072026-PSA, July 20, 2026. Warns that AI-generated deepfake videos of a senior FBI leader, hosted on a social media platform, are being used to promote spoofed IC3 lookalike sites and re-victimize prior scam complainants via Facebook Messenger and Telegram.
Scammers Impersonating Law Enforcement and Government Officials in Fraud Schemes. FBI Internet Crime Complaint Center. Public Service Announcement I-091726-PSA, September 17, 2026. Update to the March 2022 PSA reporting 60,940 complaints and $1.6B in losses (Jan 2025 - July 2026), with AI-generated video calls named as one tradecraft vector. Sub-categories: jury-duty scams (6,833 complaints, ~$36M), license-expiration scams (496 complaints, ~$348K), medical-license scams (3,322 complaints, ~$37M), foreign-targeted scams (1,809 complaints, ~$140M).
Blackburn, Coons Bipartisan Bill to Protect Individuals and Creators from Deepfakes Passes Senate Judiciary Committee. Office of Senator Marsha Blackburn. June 18, 2026. Confirms committee passage of the bipartisan bill; the first federal legislative action of 2026 on political deepfakes.
China Aims A.I. at Predicting Who Could Pose a Political Risk. The New York Times. June 1, 2026. Reports Vanderbilt's follow-up to the GoLaxy documents, showing the same AI-profiling infrastructure is now being used to identify and target Chinese citizens.
AI campaign deepfakes are starting to draw legal threats. Axios. October 2, 2026. Reports that Rep. Derrick Van Orden (R-Wis.) received a cease-and-desist letter over an AI deepfake he posted falsely depicting his Democratic opponent; the first 2026 election-cycle instance of a sitting member of Congress on the receiving end of a deepfake-related legal demand. Notes that as of June 2026, 31 states have election deepfake laws.