Deepfakes and the integrity of evidence in family court.

A custody hearing turns on credibility, and nothing wrecks credibility faster than a parent shown on screen screaming threats at a child. So picture the parent who never said those words. A short clip arrives in the file. It looks real, it sounds real, and it lands before anyone has tested where it came from. The damage is done at the moment of viewing. Unwinding it is slow, expensive, and uncertain.

That is not a hypothetical. In a 2020 English family case, a mother used consumer software and online tutorials to doctor an audio recording so that the father appeared to make violent threats; the fabrication was caught only because his lawyers obtained the original file and examined its metadata.^[1] The case is now a fixture in legal commentary precisely because it shows how little skill the attack requires and how much luck the defense needed.

The rules already contemplate this — most lawyers just don't invoke them

The instinct is to treat deepfakes as a gap in the law. They aren't. Federal Rule of Evidence 901(a) has always required a proponent to “produce evidence sufficient to support a finding that the item is what the proponent claims it is.”^[2] That standard does not assume good faith. It assumes nothing. A video is not self-proving merely because it plays.

Rule 901(b)(9) is the underused tool here. It authenticates an item by “evidence describing a process or system and showing that it produces an accurate result.”^[2] Applied to a phone video, that means the proponent should be able to account for the device, the capture, and the unbroken path from recording to exhibit — not just assert that the clip is genuine. The companion self-authentication rules, 902(13) and 902(14), let a qualified person certify that an electronic record is the accurate output of a system, or that a copied file matches its source by hash value.^[3] Used affirmatively, they reward the party who can show provenance and quietly expose the party who can't.

The burden is the battlefield

In practice the fight is rarely about whether deepfakes exist. It is about who has to prove what, and at whose cost. When a clip with no provenance is offered, the targeted parent is pushed into proving a negative — that an event never happened — which usually means retaining a digital-forensics expert to analyze compression artifacts, metadata, and edit history. Few family-court litigants can afford that, and the supply of qualified examiners is thin.^[4]

The corrective is to keep the burden where Rule 901 puts it. A court that demands a genuine foundation before a video is shown — source, custody, and a process account — does more to protect a custody record than any detector. Authentication is a threshold the proponent must clear, not a defense the respondent must fund.

The liar's dividend cuts the other way too

There is a second hazard, and it is the mirror image of the first. Once everyone knows video can be faked, the guilty parent can wave away a real recording as “probably a deepfake.” Legal scholars call this the liar's dividend: the mere existence of synthetic media lets bad actors discredit authentic evidence.^[4] A custody court that grows reflexively skeptical of all video does not become safer. It becomes blind in both directions — admitting fakes and discarding truth.

This is why provenance beats suspicion. A clip that was signed at capture and carried through an intact chain of custody can be credited with confidence. A clip with no such record gets the scrutiny it has earned. The goal is not to distrust video. It is to distrust unverifiable video.

What's coming, and what to do now

The Advisory Committee on Evidence Rules has published a proposed Rule 707, which would subject machine-generated evidence offered without a sponsoring expert to the reliability requirements of Rule 702 — closing the gap where a party tries to launder an algorithm's output past the standards a human expert would have to meet. The proposal went out for public comment through February 2026, with the committee's review continuing this year.^[5] Separately, states have moved fast on synthetic media: the National Conference of State Legislatures tracks dozens of deepfake bills, with deceptive audio and visual media among the most active categories.^[6] The statutory frame is forming around the courts even as the rules catch up.

None of that helps the parent in next week's hearing. Three habits do, and they are available today:

1. Demand a foundation before the clip plays. Under Rule 901(a), make the proponent account for source and custody first. A video that cannot survive that question should not be shown to the finder of fact.^[2]
2. Preserve and pull the original. The 2020 UK father was saved by metadata on the original file.^[1] The exported, re-compressed copy a party hands over is not the evidence — the source file is.
3. Use provenance affirmatively. Where your own evidence is signed at capture or hashable to its source, certify it under 902(13)/(14) and let the contrast with the unverifiable clip speak.^[3]

The lesson is procedural, not technical. The technology that fabricates a custody video will keep improving; the discipline that keeps it out of a child's case is older than the technology, and it is already written into the rules.