Deepfakes as evidence: how do litigators authenticate media now?

When a client hands over a recording that could decide a matter, or opposing counsel produces one that could sink it, the first question is now whether it is real. Courts have not written a new rule for deepfake evidence, so the answer depends on the existing authentication rules and on how well your firm handled the file from the day it arrived.

Is there a federal rule of evidence for deepfakes?

Not yet: no federal rule of evidence is written specifically for deepfakes. Federal courts still authenticate audio, video and images under the general standard in Federal Rule of Evidence 901(a), which asks only for evidence "sufficient to support a finding that the item is what the proponent claims it is." The federal Advisory Committee on Evidence Rules has been studying a possible deepfake amendment, and in its May 2026 report to the Standing Committee it concluded that an amendment to Rule 901 to address deepfakes "is not warranted" for now.

The report gives the committee's reasons. A Federal Judicial Center survey found that only 15 of 931 responding judges had dealt with deepfake issues, and the Department of Justice did not believe the draft was ready for publication. The committee kept a working draft in reserve and planned to vet it at its fall 2026 meeting with AI experts and litigators. A separate proposal, Rule 707 on machine-generated evidence, was published for comment in August 2025, and the same report states that the committee "does not recommend action on the proposed Rule 707 at this time."

Any new rule would still have to go through public comment, Judicial Conference approval, Supreme Court adoption and congressional review before taking effect. State courts, including California's, apply their own evidence codes, so litigators should check each forum's current rules and standing orders.

How are audio and video evidence authenticated under Rule 901?

Audio and video are authenticated under the same rules as any other evidence, with the proponent choosing from the illustrations in Rule 901(b) and the self-authentication provisions in Rule 902. The common routes are:

• A witness with knowledge testifying that the item is what it is claimed to be, under Rule 901(b)(1). For a recording, that is usually someone who was present or who recognizes the voices.
• Distinctive characteristics, meaning "the appearance, contents, substance, internal patterns, or other distinctive characteristics of the item, taken together with all the circumstances," under Rule 901(b)(4).
• Evidence describing a process or system and showing that it produces an accurate result, under Rule 901(b)(9).
• A certification of a qualified person for a record generated by an electronic process or system, or for data copied from a device and authenticated by a process of digital identification such as a hash value, under Rules 902(13) and 902(14). The proponent must give reasonable written notice so the other side has a fair opportunity to challenge the record.

The threshold is deliberately low. Once the proponent clears it, disputes about authenticity usually go to the weight of the evidence rather than its admissibility. That is the gap deepfakes exploit: a convincing fabrication with a confident sponsoring witness can clear Rule 901(a), leaving the opposing party to fight it in front of the jury.

What did the draft deepfake amendment propose?

The working draft proposed a two-step framework that shifts the burden once a credible challenge is made. Under the draft described in the committee's report, a party challenging an item as fabricated "must present evidence sufficient to support a finding of fabrication to warrant an inquiry by the court." If that showing is made, the item would be admissible "only if the proponent demonstrates to the court that it is more likely than not authentic."

The draft is not law and may never be adopted in that form. It is still a useful map of what a judge is likely to want to hear in a contested authenticity dispute: a concrete basis for suspecting fabrication rather than a general claim that anything could be AI, and a proponent who can account for where the file came from and how it was handled.

Can deepfake detection software settle whether a file is fake?

No, deepfake detection software cannot settle authenticity on its own. NIST's report on synthetic content, NIST AI 100-4, describes provenance tracking, watermarking, metadata recording and synthetic content detection as separate technical approaches, each with limits. Detection tools return a probability rather than a finding, and their results can be contested like any other expert opinion.

For litigators, the practical consequence is that the strongest authentication evidence is usually about provenance: the original device or account, the native file with intact metadata, a hash value recorded at collection, and a documented chain of custody. A screenshot of a video or a re-encoded file forwarded through a messaging app strips out most of what an examiner would need.

What should a firm preserve when a client hands over a recording?

A firm should preserve the original recording in its native form, with a record of where it came from. The duty to preserve electronically stored information under Federal Rule of Civil Procedure 37(e) already reaches this material once litigation is reasonably anticipated, and losing the original file can leave the firm unable to answer a fabrication challenge or a spoliation motion.

A defensible intake record captures who provided the file, on what date, from which device or account, and how it was transferred to the firm. It records a cryptographic hash value at intake so the firm can later show the file has not changed. It stores the original in a restricted matter location with access logging, and it works only from copies. If the source is a phone or a cloud account, the firm should consider whether a forensic collection is warranted before the data rolls off.

The California Rules of Professional Conduct tie this to competence. Comment 1 to rule 1.1 includes the duty to keep abreast of "the benefits and risks associated with relevant technology," and synthetic media is now part of that technology.

Are deepfakes only an evidence problem?

No. The same tools are used against law firms directly. A cloned voice of a partner or client asking for an urgent wire, or a video call with a fabricated participant, is a payment fraud problem before it is ever an evidence problem. Our guide to verifying trust account payouts covers the call-back controls that address these attempts.

What should a litigation practice do this quarter?

• Write a one-page intake procedure for client-provided audio, video and images: native files only, source recorded, hash value captured, original stored read-only.
• Confirm where evidence files live in your document management system or Microsoft 365, who can reach them, and whether access is logged. Our article on litigation holds in Microsoft 365 covers the preservation side.
• Identify a qualified forensic examiner before you need one, so a fabrication challenge does not start with a search for an expert.
• Add a deepfake scenario to the firm's wire verification training and confirm every payment change is verified by a call to a known number.
• Assign someone to track the Advisory Committee's work on Rule 707 and the deepfake draft, and any standing orders in the courts where the firm appears most often.

How does s90 help?

s90 manages the systems a law firm relies on to keep evidence intact: restricted matter storage, access and audit logging, retention and litigation holds in Microsoft 365, and the controls that stop voice and email fraud from reaching the trust account. We can help you set up an intake location for client media that records who touched each file and when, so your chain of custody does not depend on memory. If that would be useful, we are glad to talk it through.