The Pulse
U.S. Courts Weigh Burden Shift for Deepfake Evidence
Federal evidence officials are considering a new authentication standard for evidence challenged as AI-generated or altered. The proposal would require an initial showing from the opponent before shifting the burden to the evidence’s propon
AI.info Team ·
Federal court rulemakers are weighing a new way to handle evidence that one side claims is a deepfake, but the proposal has not cleared the committee process and officials are still divided over how quickly the rules should change.
The possible amendment to Federal Rule of Evidence 901 would create a two-stage authentication test. An opponent would first have to present enough information for a reasonable person to find that an item was fabricated or altered with generative artificial intelligence. If that threshold were met, the burden would shift to the party offering the evidence, which would then have to show by a preponderance of the evidence that the item was authentic.
The proposal appears in the September 2026 agenda book for the Judicial Conference’s Advisory Committee on Bankruptcy Rules. The document summarizes work by the Advisory Committee on Evidence Rules and says the deepfake issue remains under consideration rather than adopted as a rule.
Fifteen Judges Have Faced Deepfake Challenges
The Federal Judicial Center surveyed federal judges at the request of the Evidence Rules Committee. The September agenda materials say 15 judges reported encountering a deepfake challenge, while a strong majority said they would require a showing similar to the proposed threshold even without a formal amendment.
The committee’s concern is that the existing authenticity standard may not give judges enough structure when a party alleges that an image, recording or video has been generated or altered by AI. The proposed approach would not make every digital exhibit subject to forensic examination. A party would first need to provide evidence supporting the claim that the material may be synthetic or manipulated.
That distinction matters because the committee is also trying to prevent unsupported objections from delaying trials. A general assertion that deepfakes are easy to create would not, by itself, satisfy the proposed test.
Rulemakers Pull Back on AI Evidence Proposal
The deepfake discussion is proceeding alongside work on a possible new Rule 707 concerning machine-generated opinions offered without an expert witness. The Evidence Rules Committee received 59 written comments and testimony from about a dozen witnesses at two hearings after publishing the proposed rule.
Feedback on Rule 707 was mixed. Comments raised concerns about the breadth and vagueness of the language, its relationship with Rule 702 governing expert testimony, and whether a new rule was necessary or premature. Some participants also questioned whether the civil and criminal rules should require advance disclosure when parties plan to offer AI-generated opinion evidence.
The committee revised the proposal in response to the comments but then chose not to move it forward immediately. Professor Daniel J. Capra, the committee’s reporter, said most practicing lawyers were misunderstanding the rule, according to the September agenda materials. The committee decided that further work with technical experts and experienced lawyers would be more useful than advancing language that participants had interpreted inconsistently.
“Most of the practicing lawyers were misunderstanding the rule.”
Professor Daniel J. Capra, reporter for the Advisory Committee on Evidence Rules
Technology Moves Faster Than Federal Rulemaking
The committee’s hesitation reflects a direct conflict in the record. Judges and lawyers want a clear procedure before synthetic evidence becomes a routine trial issue, but a rule written for current AI systems could become outdated during the years required for federal rulemaking.
The agenda materials say Judge Jesse M. Furman, chair of the Evidence Rules Committee, raised concern that technology was moving quickly enough to leave any new rule behind. A lawyer member argued that the judiciary should not wait until the problem becomes unbearable before responding. The committee ultimately chose additional information gathering.
The next step is a mini-conference involving technical experts and experienced lawyers. The event is intended to inform work on both the proposed deepfake amendment to Rule 901 and the proposed Rule 707 on machine-generated opinions.
New York Case Shows the Dispute Is Already in Court
The federal committee’s review comes as state courts confront similar disputes. The agenda book records a discussion between Judge Furman and Professor Capra about a recent New York Court of Appeals decision involving deepfake evidence.
That decision, Matter of M.S. (M.H.), described the same basic burden-shifting structure under consideration by federal rulemakers: an opponent must provide enough information for a reasonable person to find that evidence was fabricated in whole or in part with generative AI, after which the proponent must show that it is more likely than not authentic.
The New York opinion also warned against treating a bare accusation as enough to force the evidence’s sponsor into an expensive authentication exercise. The ruling described a broad claim that an item is a deepfake as insufficient without a foundation in the record.
No New Federal Rule Yet
The federal proposal remains a working draft. It has not been approved by the Evidence Rules Committee, the Standing Committee on Rules of Practice and Procedure, the Judicial Conference, the Supreme Court or Congress.
The current record points toward a narrower intervention than a blanket requirement for expert testimony on every digital recording. Under the approach being studied, the party alleging AI manipulation would have to establish a factual basis first. Only then would the proponent face the higher burden of proving authenticity by a preponderance of the evidence.
For now, federal judges continue to apply the existing evidence rules while the committee gathers more technical and legal input. The September 23, 2026 agenda materials leave the central question open: whether a new rule can remain precise enough to help courts without becoming obsolete before it takes effect.