Deepfakes In Personal-Injury Litigation – Authenticating Digital Evidence Without Rewarding Speculation

The Problem Is Authenticity, Not Just AI

Video often shapes disputes over liability, timing, impact, impairment, and injury. Dashcam, surveillance, and cellphone footage can be decisive in personal-injury litigation. Generative AI complicates their use. Fabricated recordings may appear genuine, while authentic footage may be challenged as a deepfake without supporting evidence. Courts must distinguish credible authenticity disputes from speculation.

The central questions are whether the recording is properly authenticated, whether specific evidence justifies further examination, and what inquiry is proportionate. Disclosed AI-generated demonstratives should be distinguished from recordings allegedly altered or fabricated without disclosure.^[1]

The Governing Framework

Under Federal Rules of Evidence 901(a) and 104(b), the proponent must present evidence sufficient to support a finding that the exhibit is what the proponent claims it to be. Rule 901(b) permits foundations including knowledgeable-witness testimony, distinctive characteristics, and evidence about a reliable process or system. Authentication is a threshold issue, not proof every depicted assertion is true. Hearsay, relevance, prejudice, and original-or-duplicate rules may independently matter. ^[2]

Rules 902(13) and 902(14) permit self-authentication of certain electronic-process records or digitally identified copies when certification requirements are met. They can help establish process or file integrity, but do not alone prove that a video depicts the claimed event. A disclosed AI animation raises different issues from a video offered as an unaltered recording.

What the Decisions Actually Establish

In Mendones v. Cushman & Wakefield, Inc., the Alameda County Superior Court imposed terminating sanctions in September 2025 after finding that self-represented plaintiffs intentionally submitted false evidence in support of summary judgment. The court identified purported witness videos bearing characteristics of AI-generated content and questioned other exhibits. The ruling illustrates the consequences of fabricated submissions, not a nationwide requirement for forensic review of every video.^[3]

In Matter of M.S. (M.H.), the New York Court of Appeals ruled that videos recovered through a third party lacked sufficient authentication. The court reversed the appellate orders and dismissed the child-protective petitions without finding that the recordings were deepfakes. The dissent argued that the majority imposed excessive authentication burdens despite the absence of a fabrication claim below. The decision turns on New York law and the particular gaps in the evidentiary foundation.^[4]

In Frazier v. Southeast Georgia Health System, Inc., the Eleventh Circuit affirmed dismissal after the district court found a purported examination-room video fabricated; it was not an AI-deepfake case. In a 2023 tentative discovery ruling in Sz Hua Huang v. Tesla, Inc., the court rejected a generalized deepfake-based refusal to admit or deny statements attributed to Elon Musk and allowed a limited deposition about identified recordings. That ruling concerned discovery, not final admissibility. Fabrication allegations require evidence, and “deepfake” is not a talismanic objection. ^[5]

Rulemaking Remains Unsettled

As of October 9, 2026, the Federal Rules of Evidence contained no dedicated deepfake-authentication provision. In May, the Evidence Rules Advisory Committee concluded that amending Rule 901 was not warranted at that time, while continuing work on possible Rule 901(c) language. A separate Rule 707 proposal addressing certain machine-learning evidence also remained under study. The October 15 meeting agenda scheduled further discussion; neither proposal was then binding law.^{[6] [7]}

A Practical Protocol for Personal-Injury Counsel

One must preserve the original promptly, and record the source device or system, collector, acquisition date and method, transfer history, and anyone who accessed or processed the file. One should preserve the native file, document all exports or conversions, and calculate a cryptographic hash where appropriate. A hash can show that a file has not changed since hashing; it cannot prove that the content was authentic initially.

For opposing footage, one should request the original or native export, relevant metadata, capture system, custody history, and details of editing, compression, enhancement, or AI tools. For surveillance, one should identify the investigator and collection process. One must tailor discovery to the issue instead of reflexively demanding source code or model weights.

A serious challenge should identify concrete anomalies, such as custody gaps, inconsistent timestamps, audio-video desynchronization, discontinuities, unexplained processing, or conflict with independent evidence. These are investigative leads, not proof of fabrication. Missing metadata may have an innocent explanation, and neither visual inspection nor automated detection alone conclusively establishes manipulation. Where justified, a digital-forensics expert should examine the source, explain methods and error limits, consider alternatives, and distinguish file integrity from scene authenticity.

If needed, one must seek a focused pretrial process under Rules 104 and 901, considering Rule 403 and original-or-duplicate rules as relevant. One should ask the court to define the dispute and choose a proportionate remedy, such as added foundation, limited examination, a forensic report, exclusion, or a limiting instruction. Both sides should disclose and label AI-assisted enhancement and preserve the untouched source. ^[8]

One shouldresolve deepfake disputes through evidence, and not assumption, must preserve provenance, investigate specific concerns, and use forensic expertise when justified preventing fabricated media from misleading the factfinder without allowing unsupported suspicion to discredit authentic recordings.

^[1] Maura R. Grossman & Paul W. Grimm, “Judicial Approaches to Acknowledged and Unacknowledged AI-Generated Evidence,” 26 Science & Technology Law Review, no. 2 (2025). Sources: Journal article / DOI; Publication record.; Abhishek Dalal et al., “Deepfakes in Court: How Judges Can Proactively Manage Alleged AI-Generated Material in National Security Cases,” 2024 University of Chicago Legal Forum 75 (2025). Source: University of Chicago Legal Forum article; Daniel J. Capra, “Deepfakes Reach the Advisory Committee on Evidence Rules,” 92 Fordham Law Review 2491 (2024). Sources: Fordham Law Review article; Publication record.

^[2] Fed. R. Evid. 901, 902(13)–(14), 104, 403, 1001–1003. Sources: Rule 901; Rule 902; Rule 104; Rule 403; Rule 1001; Rule 1002; Rule 1003; Paul W. Grimm, Daniel J. Capra & Gregory P. Joseph, “Authenticating Digital Evidence,” 69 Baylor Law Review 1 (2017). Sources: Fordham archive record; Duke scholarship record.

^[3] Cal. Code Civ. Proc. § 128.7; Cal. Evid. Code §§ 1400–1401. Sources: Code of Civil Procedure § 128.7; Evidence Code § 1400; Order re Terminating Sanctions, Mendones v. Cushman & Wakefield, Inc., No. 23CV028772 (Cal. Super. Ct., Alameda Cnty. Sept. 9, 2025) (Kolakowski, J.).

^[4] Matter of M.S. (M.H.), 2026 NY Slip Op 00825 (N.Y. Feb. 17, 2026).

^[5] Tentative Ruling on Plaintiffs’ Motion to Compel Further Responses, Sz Hua Huang v. Tesla, Inc., No. 19CV346663, at 28–29 (Cal. Super. Ct., Santa Clara Cnty. Apr. 27, 2023). Source: Tentative ruling; Frazier v. Southeast Georgia Health System, Inc., No. 24-10976, slip op. at 4–6 (11th Cir. Oct. 1, 2024) (per curiam) (unpublished). Source: Eleventh Circuit opinion.

^[6] Advisory Committee on Evidence Rules, Report to the Standing Committee, at 11 (May 17, 2026). Source: Official May 2026 report; Advisory Committee on Evidence Rules, Agenda for Committee Meeting, at 5–24 (Oct. 15, 2026). Source: Official October 2026 agenda; See Capra, “Deepfakes Reach the Advisory Committee on Evidence Rules”, supra note 3.

^[7] John P. LaMonaga, “A Break from Reality: Modernizing Authentication Standards for Digital Video Evidence in the Era of Deepfakes,” 69 American University Law Review 1945, 1984 (2020). Sources: American University Law Review article; Full-text PDF.

^[8] See Fed. R. Evid. 901, 902(13)–(14), 104, 403, 1001–1003, supra note 4; See also Grimm, Capra & Joseph, “Authenticating Digital Evidence”, supra note 6; and Grossman & Grimm, “Judicial Approaches to AI-Generated Evidence”, supra note 1.