New Court Case Exemplifies Evidentiary Challenges with Deepfakes
Editor’s Note: AI-generated video is making a basic evidentiary question increasingly more difficult to answer: Is this evidence real? In this article, HaystackID® contributing author Phil Favro examines United States v. Belt, a recent case involving disputes over the authenticity and admissibility of surveillance video. The case points to two related problems for courts: fabricated evidence presented as legitimate evidence, and a tactic known as the “Liar’s Dividend.” These questions are also before federal rulemakers, who are considering changes to the Federal Rules of Evidence that could affect how courts evaluate suspected deepfakes. Favro explains what Belt means for lawyers today, including the importance of preserving original files, documenting provenance and recording technology, maintaining chain of custody, and preparing fact-based challenges to questionable media. The case shows why lawyers may need to rethink how they authenticate, preserve, and challenge video evidence as AI-generated content becomes harder to identify.
New Court Case Exemplifies Evidentiary Challenges with Deepfakes
By Phil Favro, Contributing Author for HaystackID
The legal profession has seen sweeping changes since the advent of generative artificial intelligence (AI) and its proliferation into law practice. Several of those changes are positive. Lawyers have suggested that AI has increased their efficiency, expedited results for clients, and enhanced their overall practice acumen.
Of course, there is a flipside to AI. Lawyers have unwittingly cited phony court decisions in briefing, and experts have relied on fake authorities. Beyond hallucinations, however, is a growing phenomenon that threatens the very integrity of legal proceedings: fabricated AI-generated evidence. Although these fictions could take many forms, the most commonly discussed at this time is fraudulent video footage or “deepfakes.”
Deepfakes have been around for years. Once upon a time, they were easy to spot. The footage often had technical glitches, the audio quality was suspect, and the images themselves were just not believable. So much of that has changed with advances in AI technologies. The consensus now is that “deepfakes are easy to generate and difficult to detect.”[1]
The enhanced quality, ease of generation, and proliferation of deepfakes present unique evidentiary issues that judges and lawyers are being forced to address. One such issue includes admission of deepfakes into evidence, i.e., the potential for a party to pass off fabricated video footage as legitimate evidence. The presentation of fake evidence threatens the integrity of trials and other evidentiary proceedings. A related issue involves accusations that authentic evidence is actually fabricated. Referred to as the “Liar’s Dividend,” allegations of fraud can sow doubts and taint perceptions about legitimate evidence, particularly among jurors.
A recent court case—United States v. Belt—highlights both of these issues. In Belt, the trial court considered the authenticity of home video surveillance footage, including accusations of tampering by the government. Although admitted into evidence, questions remained on appeal regarding the authenticity of the video footage. While the appellate court ultimately dismissed those questions, dissenting judges questioned the admissibility of the footage, with one judge pushing for a retrial to address lingering authenticity concerns.
The unresolved concerns from Belt demonstrate why rulemakers are presently considering changes to address issues with deepfakes. Belt also emphasizes the need to consider practices for handling questions over the admissibility and weight of authentic and fabricated evidence.
Belt: The Crime and the Surveillance Video
Belt involved a violent attack, followed by a homicide, on a South Dakota Indian reservation. The government accused two individuals of murdering an acquaintance and assaulting his siblings. There were several pieces of evidence connecting the defendants to the crimes. They included eyewitness testimony, a DNA sample, and the victim’s blood on a vehicle driven by the defendants. There was also surveillance video from a Ring camera belonging to the victim’s aunt and on property adjacent to the crime scene.[2] The video appeared to capture several aspects of the crime, such as shouting, the assaults, a fire on the victim’s property, the defendants’ vehicle, and the eventual arrival of law enforcement. On the audio feed, a male voice is repeatedly shouting “stop” while another voice can be heard yelling “Get him, get him. Run him over. Let’s go.” The video also included several gaps ranging from seconds to over 22 minutes. During one of the gaps, a car ran over the victim.
The government obtained the video from the victim’s aunt, with an FBI agent downloading footage directly from her smartphone.
The Trial
At trial, the government sought to admit pertinent footage from the surveillance camera. The defendants objected, arguing the video could not be properly authenticated given the gaps in the recording and that the victim’s aunt was unavailable to explain the basis of those gaps. Nevertheless, the trial court overruled the objection and admitted the footage.
The video was a key aspect of the trial. As a dissenting appellate judge later observed, “[t]he video played a significant role in the government’s case at trial, where the jury watched it, multiple witnesses discussed it, and the government relied on it repeatedly in its closing.” The jury returned a verdict against the defendants, finding them guilty of second-degree murder. The court sentenced them to 22 years and 27 years in prison.
A Divided Appellate Panel Rejects Video Tampering Allegation
The defendants appealed their convictions, arguing that the trial court erred by admitting the video compilation. In support of their appeal, the defendants advanced a contrary theory to the government’s narrative about the video: one of the victim’s siblings (Sara) ran him over with the car and the aunt “deleted critical pieces of footage to protect” Sara.
The U.S. Court of Appeals for the Eighth Circuit rejected the defendants’ arguments and affirmed their convictions.[3] A divided three-judge panel concluded that the totality of the circumstances surrounding the footage confirmed that it was authentic.
In particular, the majority found that the government met the “low bar” for authentication by satisfying Federal Rule of Evidence 901(b)(4). Rule 901(b)(4) allows a party to authenticate evidence through the “distinctive characteristics” of the piece of evidence: “The appearance, contents, substance, internal patterns, or other distinctive characteristics of the item, taken together with all the circumstances” satisfy Rule 901(a)’s authentication requirement that the proponent offer “evidence sufficient to support a finding that the item is what the proponent claims it is.” The majority reasoned that the video’s contents aligned circumstantially with eyewitness testimony, e.g., matching vehicle descriptions, the fire on the victim’s property, shouting voices, the arrival of law enforcement, and the overall timeline of events surrounding the crime.
The majority likewise rejected the defendants’ alternate theory that the aunt tampered with the Ring footage. Reasoning that it was “improbable” that the aunt deleted the footage given the random nature of the gaps, the majority also emphasized that gaps in recordings—standing alone—do not render the evidence inadmissible. Instead, “[g]aps in a recording affect the weight of the evidence, not its admissibility.” (internal quotations omitted)
Dissents and Concerns Over Deepfakes
Despite affirming the defendants’ convictions, the majority did so over the dissent of the Honorable Jane Kelly, an Eighth Circuit judge who opined that the trial court erred in admitting the Ring footage.
Judge Kelly’s Dissent
Judge Kelly criticized the majority for glossing over the circumstances surrounding the video in Belt, which differed from traditional instances where law enforcement testify about footage they generated. In Belt, the FBI agent did not create the footage, but downloaded it directly from the aunt’s Ring camera. In so doing, the agent did not obtain information about the camera make or model, the software application used, or whether the system recorded continuously or was motion activated. Without that information, the government failed to offer evidence supporting the conclusion that the camera was operating as designed.
For Judge Kelly, all of this undermined the notion that the gaps in the video were inconsequential. From her perspective, the gaps impacted the video’s admissibility and not merely its weight. Given the central role of the video in the trial, Judge Kelly would have vacated the defendants’ convictions and ordered a new trial so the government could squarely address these admissibility issues.
Judge Erickson’s Dissent
After the majority rejected their appeal, the defendants subsequently petitioned for a rehearing before the three-judge panel, along with en banc review by the Eighth Circuit. The appellate court denied both requests, but not without eliciting another dissent, this time from a different Eighth Circuit judge, the Honorable Ralph Erickson.[4]
In his dissent, Judge Erickson underscored the same themes that Judge Kelly highlighted in her dissent. In addition, Judge Erickson called out the issue of deepfakes, observing that the “changing technological environment in which we live” calls for a stricter gatekeeping function by trial courts given the reality of AI-generated videos.
Federal Rulemakers Consider the Issues
The lingering evidentiary questions in Belt highlight a systemic evidentiary challenge arising from deepfake evidence: the generally low threshold for authentication under Rule 901(a) and Rule 104(b). While the flexible “sufficiency” standard has traditionally sufficed, the Federal Advisory Committee on Evidence Rules has recognized that it may be insufficient to address deepfakes. Accordingly, the Advisory Committee has been actively studying possible changes to the Federal Rules of Evidence.[5]
One proposal—pertinent to the issues raised in Belt—would impose a threshold condition that requires a party challenging a possible deepfake to first produce evidence that would support a finding of fabrication. If the condition is satisfied, the proponent of the evidence would then have to establish by a preponderance of the evidence that the video is authentic. All of this would be done outside the presence of the jury.
Styled as proposed Rule of Evidence 901(c), this suggested provision could potentially address the issues posed by deepfakes and exemplified in Belt. It could provide an enhanced gatekeeping function to better ensure that parties do not pass off deepfakes or tampered videos as legitimate evidence. In addition, it could address the Liar’s Dividend, with accusations of fraud regarding video footage—like those the defendants made in Belt—presented before the judge for determination and not to the jury.
Practical Guidance in the Meantime
While this proposal and others[6] provide a roadmap for the future, rulemakers are intentionally taking a measured approach; studies about possible rules changes are ongoing without a definite end date. Because final rule amendments remain years away from implementation, counsel should consider certain practices for handling deepfake issues in the meantime, including drawing upon existing evidentiary rules.
Evidence Proponents
Parties seeking to introduce media (including video footage and audio feeds) as evidence may need to consider the possibility of deepfake challenges by adversaries. To better establish authenticity, parties may need to show the provenance of the video. One way to do so is by collecting and preserving original native files rather than relying on informal file transfers like the download that took place in Belt.
Another potentially useful step is to document the make and model of the recording hardware, the specific software application or firmware version used, and the operational parameters (such as whether the device records continuously or uses motion-activation sensors). While Belt ultimately did not require this information, a different judge—as the Belt dissents make clear—could very well direct the proponent to supply such information for authenticity purposes.
Belt also makes clear that proponents should consider creating a chain of custody log that tracks transfers of the evidence in question. In addition, proponents should be prepared to proactively address gaps in video footage or audio feeds. Expert witness testimony and technical explanations may be needed to bolster fact witness testimony.
Evidence Opponents
Parties opposing the introduction of deepfake videos should ground their accusations in fact-specific evidence. This may very well require the retention of experts to evaluate the quality and nature of the recording at issue. In addition, the opponent (and its expert) may need to inspect the underlying file structure for the video to assess any number of issues, including whether the application permits media editing and whether the footage or feed was in some way manipulated.
Finally, opponents should consider drawing upon current evidence rules to challenge deepfake evidence. This could include seeking to have the court (rather than a jury) address the admissibility of media under Rule 104(a). Opponents could also invoke the protections of Rule 403, particularly factors such as “unfair prejudice, confusing the issues, [or] misleading the jury,” to persuade the court not to admit the questionable footage or feed.[7]
[1] Report of the Advisory Committee on Evidence Rules, at 11 (May 17, 2026), available at https://www.uscourts.gov/sites/default/files/document/advisory_committee_on_evidence_rules_may_2026.pdf. [2] See United States v. Belt, No. 3:23-cr-30067-RAL (D.S.D. June 28, 2024), ECF No. 149 (Exhibit List) (referencing Exhibit 184 as reflecting “Ring security camera video – compilation”). [3] United States v. Belt, 175 F.4th 993 (8th Cir. 2026). [4] United States v. Belt, 181 F.4th 931 (8th Cir. 2026). [5] See generally Report of the Advisory Committee on Evidence Rules (May 17, 2026), available at https://www.uscourts.gov/sites/default/files/document/advisory_committee_on_evidence_rules_may_2026.pdf. [6] The Advisory Committee is also considering a proposal—styled as proposed Rule 707—that would impose gatekeeping requirements for acknowledged AI-generated evidence. See id., at 6-11; Maura R. Grossman & Hon. Paul W. Grimm (ret.), Judicial Approaches to Acknowledged and Unacknowledged AI-Generated Evidence, 26 Colum. Sci. & Tech. L. Rev. 110, 120-21 (2025). [7] See generally Grossman & Grimm, Judicial Approaches to Acknowledged and Unacknowledged AI-Generated Evidence (discussing the factors affecting the use of Rule 104(a) and Rule 403 to challenge the admissibility of deepfakes).
About Phil Favro
Phil Favro is the founder of Favro Law PLLC, where he counsels clients on ESI, AI, and discovery issues and serves as a special master, mediator, and expert witness. Phil is nationally recognized for his expertise on ESI, discovery, and information governance, with courts acknowledging his credentials. See, e.g., Oakley v. MSG Networks, Inc., No. 17-CV-6903 (RJS), 2025 WL 2061665 (S.D.N.Y. July 23, 2025). This background makes Phil particularly well-suited to counsel clients and advise courts on information-related issues. As a special master, Phil is acclaimed for his collaborative approach, working with parties to find stipulated solutions to complex issues. For disputes that require adjudication, he is renowned for the clarity and vigor of his written dispositions, which are available on legal search engines.
About HaystackID®
HaystackID® solves complex data challenges related to legal, compliance, regulatory, and cyber requirements. Core offerings include Global Advisory, Cybersecurity, Core Intelligence AI™, and ReviewRight® Global Managed Review, supported by its unified CoreFlex™ service interface and eDiscovery AI® technology. Recognized globally by industry leaders, including Chambers, Gartner, IDC, and Legaltech News, HaystackID helps corporations and legal practices manage data gravity, where information demands action, and workflow gravity, where critical requirements demand coordinated expertise, delivering innovative solutions with a continual focus on security, privacy, and integrity. Learn more at HaystackID.com.
Assisted by GAI and LLM technologies.
SOURCE: HaystackID