When Authenticity Becomes Contested: New Challenges for U.S. Evidence Law in the Age of AI

In the past, a photograph, an audio recording, or a video often carried immediate persuasive force in court. Witnesses may have imperfect memories or give inconsistent accounts, but people tend to trust what they can see and hear for themselves. The rapid development of generative artificial intelligence, however, is testing that long-standing intuition.

A recent California case, Mendones v. Cushman & Wakefield, Inc., illustrates the problem. The plaintiffs submitted several videos and images as evidence. During its review, the court identified unusual features, including limited facial movement, looping footage, and other irregularities inconsistent with genuine recordings. The court concluded that certain videos were AI-generated deepfakes and imposed terminating sanctions, dismissing the action with prejudice. The case did not create a new evidentiary rule for AI-generated material. It did, however, expose an increasingly practical question: when AI can produce convincing video and audio, how should a court determine whether digital evidence is authentic?

That question is becoming one of the central evidentiary challenges of the AI era.

U.S. Evidence Law Relies on Authentication, Not Appearance Alone

Many people assume that once a party produces a photograph, recording, or electronic file, the material can simply be used as evidence. That is not how U.S. evidence law works.

Federal Rule of Evidence 901 requires the proponent to authenticate an item as a condition of admissibility. In practical terms, the proponent must produce evidence sufficient to support a finding that the item is what the proponent claims it is.

Courts therefore do not ask only whether evidence looks genuine. They also examine how it was created, obtained, stored, and preserved. A photograph may be authenticated by a witness with knowledge of the scene or the circumstances in which it was taken. An audio recording may be authenticated through testimony about the recording process, the handling of the file, and the identity of the speakers. An email may be authenticated through its contents, transmission information, metadata, server records, or surrounding circumstances.

In other words, U.S. evidence law does not rely solely on the apparent credibility of the item itself. It relies on the evidentiary foundation that connects the item to its claimed source and history.

That framework has generally adapted to decades of technological change. Generative AI, however, is placing new pressure on it.

The Deeper Challenge Is Not Merely Fake Evidence, but the Growing Difficulty of Distinguishing Real from Fake

If Mendones shows how a court may confront apparently AI-generated evidence, a recent federal case illustrates the opposite problem.

In Burnley v. Valentin, a party relied on an audio recording, and the opposing party argued that the voice might have been created or cloned through AI. A challenge of that kind would have seemed unusual only a few years ago. It is now appearing in actual litigation.

The court did not exclude the recording merely because an AI-based fabrication was theoretically possible. Nor did it abandon the traditional rules of evidence. Instead, it returned to the ordinary authentication inquiry. The court considered sworn declarations explaining how the call was recorded, how the recording was transmitted, and whether the file before the court was the original or an exact duplicate. It also considered testimony identifying the speaker’s voice and the surrounding circumstances.

Importantly, the court rejected the argument that Rule 901 required a flawless or fully documented chain of custody. A strict chain of custody is not an absolute requirement. The relevant question is whether the proponent has produced sufficient evidence that the recording is what it purports to be and has not been materially altered.

After reviewing the evidence and conducting an evidentiary hearing, the court found that the recording satisfied Rule 901 and was authentic.

Together, Mendones and Burnley show the two directions in which AI is affecting evidence law. Courts must prevent fabricated digital material from entering the record. At the same time, genuine evidence may now be challenged simply by asserting that it could be a deepfake.

The problem created by AI is therefore not limited to the easier production of false evidence. Authenticity itself is becoming a contested issue in litigation.

U.S. Courts and Rulemakers Are Considering New Responses

The United States has not yet adopted a single, settled evidentiary framework for AI-generated or AI-manipulated evidence. Courts continue to rely primarily on the existing Federal Rules of Evidence, including Rule 901 for authentication and Rule 702 when expert testimony is used to address technical reliability.

One possible development is a more demanding preliminary process when a party presents a concrete basis for believing that audio or visual evidence was generated or materially altered by AI. Under proposals considered by federal rulemakers, the court could require a stronger showing of authenticity before allowing the evidence to reach the jury. Relevant proof might include the original file, metadata, device information, system logs, corroborating evidence, and documentation showing how the material was handled.

Digital forensics and expert testimony are also likely to become more important. In disputes involving complex audio, video, or other electronic evidence, courts may increasingly rely on forensic analysis to determine whether a file was edited, compressed, synthesized, or otherwise processed, rather than relying only on visual or auditory inspection.

Legal scholars and federal rulemakers are also considering whether the existing rules should be revised. Proposed Federal Rule of Evidence 707 would address certain AI-generated outputs that function like expert opinions but are offered without a sponsoring expert. A separate draft amendment to Rule 901(c) has focused on deepfake-related authentication disputes. As of July 2026, however, neither proposal has been adopted, and the Advisory Committee continues to study both issues.

For now, courts will likely continue to apply existing evidence rules case by case and build a more developed body of law through individual disputes, rather than immediately replacing the current framework with an entirely new system for AI evidence.

Conclusion

Technological change continually reshapes the questions that the law must answer.

Decades ago, courts focused on whether a photograph had been edited, an audio recording had been spliced, or an email had actually been sent. Today, the problem is more complex. A person shown in a video may never have appeared before a camera. A voice may reproduce words that the speaker never said. An entire scene may have been generated by an algorithm.

AI has not changed the basic objective of evidence law: searching for and determining what constitutes reliable proof of what happened. AI is, however, changing the methods by which authenticity and reliability must be established.

For lawyers, litigation may increasingly require not only finding evidence but also showing why that evidence deserves to be trusted. For businesses, preserving original files, metadata, system logs, and documented evidence-handling procedures will become an increasingly important part of risk management.

Courts have traditionally asked whether a particular item is authentic. In the age of AI, they may first need to answer a more basic question: what is sufficient to prove that an item is authentic?

References

1. Mendones v. Cushman & Wakefield, Inc., No. 23CV028772, 2025 Cal. Super. LEXIS 57811 (Cal. Super. Ct. Alameda Cnty. Sept. 9, 2025).

2. Burnley v. Valentin, No. 3:23-cv-00160, 2026 U.S. Dist. LEXIS 57133, 2026 WL 767145 (E.D. Va. Mar. 18, 2026).

3. Quinn Emanuel Urquhart & Sullivan, LLP, Adapting the Rules of Evidence for the Age of AI (Nov. 6, 2025).

4. Hon. Joseph H.L. Perez-Montes, Daubert in the Digital Age—Proposed Federal Rule of Evidence 707, Westlaw Today (June 24, 2026).