Federal Judges Set Bar for Deepfake Evidence Challenges

A “deepfake” objection backed by nothing more than the word itself will get a litigant nowhere in most federal courtrooms, according to a recent survey of federal judges and magistrates.

This conclusion emerges from a survey released March 25, 2026, by the Federal Judicial Center. The Administrative Office of the U.S. Courts’s Advisory Committee on Evidence Rules had requested the survey to inform its work on possible changes to Federal Rule of Evidence 901, which governs the authentication of evidence.

A ‘deepfake’ objection backed by nothing more than the word itself will get a litigant nowhere in most federal courtrooms.

The survey results are included on Page 226 in the meeting materials for the May 7 meeting.

From the 931 judges who responded – a 45% response rate – the survey collected views on two related questions. Should the rules change to address deepfake challenges? And what kind of initial showing should the challenger make to put the authenticity question into play?

Deepfake Challenges Remain Rare

The first headline finding may surprise readers who follow the legal community’s growing alarm over generative artificial intelligence in litigation. Only 15 of the 931 responding judges – roughly 2% – had ever encountered a litigant challenging audiovisual evidence as a deepfake. Among those 15, two-thirds had seen only one such challenge across calendar years 2024 and 2025. Most encounters arose in civil cases.

The Federal Judicial Center researchers tracked down four of the cases through PACER. One, Freedman Normand Friedland LLP v. Cyrulnik, No. 1:21-cv-01746 (S.D.N.Y. April 5, 2023), involved a dispute over whether audiovisual materials had been AI-generated or manipulated. The court rejected the supporting expert report as untimely and never reached the merits. Another, United States v. Sheffler, No. 3:19-cr-30067 (C.D. Ill. July 14, 2022), turned out to involve a “zoomed-in” clip of admitted evidence – not the kind of AI manipulation the committee aims to address.

The most striking example came from United States v. Hernandez, No. 3:25-cr-2384 (W.D. Texas, Jan. 21, 2026). A confidential informant pleaded guilty to obstruction of justice after providing false audiovisual evidence to federal prosecutors. The fabricated material had earlier produced an indictment of an innocent person whose case the government dismissed without explanation. Falsity surfaced only at the informant’s sentencing hearing.

Two takeaways follow. First, deepfake challenges have not yet flooded the federal docket. Second, the most consequential fabrication problems may never reach the docket at all.

What Should Trigger Judicial Inquiry?

Most federal judges would not let a “deepfake” objection proceed on the strength of the word alone. Of the 914 judges who had not encountered a deepfake challenge, 745 (82%) said they would demand an initial showing from the challenger before inquiring further into authenticity. Among the 15 judges with first-hand experience, the responses split nearly evenly: 8 required an initial showing, 9 did not.

When pressed on what that initial showing should look like, 679 judges supplied written descriptions. The Federal Judicial Center sorted the answers along two axes – the level of showing demanded and the nature of the attack on the evidence.

On the level of showing, the responses arrayed across a spectrum:

• Minimal evidence (42%) – the most popular answer. Judges in this group would accept an affidavit, technical evidence pointing to inauthenticity, or comparable corroboration from the challenger.
• Reasonable argument (35%) – a “good faith reasonable basis for asserting falsification,” or a reason to doubt authenticity beyond mere speculation.
• Expert evidence (3%) – a smaller group would demand qualified expert testimony at the threshold.
• Sworn testimony (3%) – a similarly small group would require sworn statements.
• Stronger showing from the proponent (4%) – some judges would shift the inquiry to the offering party, requiring a more robust authentication.
• Depends on the case (6%) – others reserved judgment, conditioning the answer on case-specific facts.

On the nature of the attack, 13% of comments described a direct factual challenge such as a “tell” in the video, a glitch, a feature that does not match. Another 8% described an indirect attack, such as an alibi: “This is a picture of him in [location]; he has never been to [location].”

The picture that emerges aligns closely with the threshold approach already drafted by the Advisory Committee on Evidence Rules. The committee’s proposed Rule 901(c) would require a challenger to present evidence sufficient to support a finding of fabrication before the court inquires further. A bare assertion that an item is a deepfake would not suffice. The survey suggests that federal judges, by and large, have already arrived at that destination on their own.

Should Rule 901 Be Amended?

Survey responses on the broader question of rule reform showed a slimmer majority. Among judges with no deepfake experience, 487 of 889 (55%) favored amending the rules; 402 (45%) opposed. Among the smaller group with first-hand experience, the numbers ran the other way: only 5 of 15 (33%) supported amendment.

The split reflects an institutional debate that has played out at the committee level for two years. As covered in reporting on the New York State Bar’s task force on artificial intelligence, some experts urge a “wait-and-see” approach, arguing that trial courts deserve room to develop case-by-case standards before the rule makers step in. The judges who have actually fielded deepfake challenges appear to share that caution. Their colleagues who have not yet seen one – and may be reasoning about what might happen – tilt slightly the other way.

What This Means for Litigators

Several practical conclusions follow for litigators preparing for the deepfake era.

First, a “deepfake” objection raised without supporting material will go nowhere in most courtrooms. Roughly four in five federal judges will demand some initial showing, and many will reject conclusory assertions on the spot. The “deepfake defense” – attacking authentic evidence as fabricated – faces stiff headwinds at the federal trial bench.

Second, the strongest objections combine the level of showing with the nature of the attack. An affidavit, technical analysis, or expert report becomes more persuasive when paired with a specific anomaly in the file or an external fact that contradicts what the file claims to depict. Litigants who can pair the two stand the best chance of triggering further judicial inquiry.

Third, the survey signals where federal judges already lean. The committee’s consideration of a burden-shifting framework reflects what most judges already say they would do under the current authentication rules. This is a useful data point for any party planning a “deepfake” challenge or defending against one.

Fourth, the rare cases that have surfaced suggest that fabrication concerns often resolve outside the formal record – through guilty pleas, government dismissals, or oral rulings never reduced to writing. Counsel who suspect manipulated evidence should consider how to create a record that survives motion practice and appellate review, even when the trial court appears ready to handle the matter informally.

The advisory committee’s May 7 meeting may not produce a final rule. The Federal Judicial Center survey results, however, give the rule makers – and the lawyers who will appear before them – the clearest map yet of how federal judges think about challenges to audiovisual evidence in the age of generative artificial intelligence.