Here is the uncomfortable position the American legal system now occupies. In a world where generative AI can produce convincing fake audio, video and photos on a laptop or phone, the only reliable way to prove a digital file is real is digital forensics, ideally tied back to the originating device and an intact chain of custody. Anything less, and a competent opposing expert can raise reasonable doubt about almost any digital exhibit.

That is the right answer. Its rightness is exactly what makes it weaponizable. Any tool this powerful needs procedural guardrails, or whichever side has the incentive to run up costs can use it to force settlements that have nothing to do with the merits.

I am already watching it happen.

The Liar's Dividend Has Two Payouts In AI Evidence Disputes

When Bobby Chesney and Danielle Citron coined "the liar’s dividend" in 2018, they described a defensive payoff. Real footage surfaces. The subject claims it is fake. Doubt does the rest. That is the version that gets written about. I covered that side of the dividend in an earlier piece on why detection tools alone are not the answer.

There is a second payout. You do not wait for inconvenient evidence to surface. You preemptively challenge the authenticity of a digital exhibit the other side plans to rely on. Cell phone video and voicemails. Surveillance footage, dashcam recordings, electornic documents, body-worn camera video. You force them to authenticate, at their expense, on your timeline.

The evidence does not have to be fake. It just has to be challengeable. In 2026, everything is challengeable, because the authentication standard that actually works can be expensive and hard to reproduce at scale.

Why The Authentication Floor Keeps Rising For Digital Evidence

Federal Rule of Evidence 901 requires only that a proponent produce evidence "sufficient to support a finding that the item is what the proponent claims it is." A custodian saying "yes, that is our surveillance footage from that day" usually cleared it.

Claims that a recording was fabricated are not new. I have heard them for years, in cases where the technology to actually fabricate convincing audio, video, or photos was well out of reach of the person making the claim. What has changed is the plausibility of the assertion. Producing convincing synthetic media used to require technical skill, time, and resources that most litigants simply did not have. That barrier has dropped considerably. Quality still varies, and forensic artifacts are often detectable, but the pool of people who could plausibly have generated a fake is now much larger than it used to be. That shift alone is enough to change how a judge hears the objection.

When counsel says "Your Honor, we cannot stipulate to authenticity given the current state of generative AI," many judges no longer treat it as frivolous. They treat it as prudent. The Judicial Conference’s Advisory Committee on Evidence Rules has now moved past the earlier proposed Rule 901(c) deepfake amendment and instead advanced proposed Rule 707, which would apply Daubert-style reliability standards to machine-generated evidence offered without an expert witness. The Advisory Committee meets May 7, 2026 to take up the rule after public comment closed February 16. Rule 707 does not solve the deepfake authentication question directly, since it applies only to evidence the proponent acknowledges is machine-generated, but it is the clearest signal yet that the federal rules are catching up, albeit slowly, to the reality litigators already face.

So the authentication standard climbs. Digital forensics becomes the practical floor. And the side that has to authenticate the most evidence pays the most.

The Cost-Weaponization Pattern Defense Counsel Is Not Talking About

The conventional version of cost-weaponized litigation assumes deep pockets belong to a corporate defendant grinding down a contingency plaintiff. That dynamic exists, and it still runs in plenty of cases. There is an inverse version that deserves more attention, because it is the one that scales fastest under current incentives.

Consider a plaintiff's case built around the claim that the defense's recorded evidence is unreliable. A defendant with clean surveillance video showing the claim is exaggerated used to be in a strong position. Now that video is a line item. Every exhibit the defense wants in front of a jury is an exhibit the other side can demand be authenticated at the device level. Each authentication fight runs up expert fees on the defense. The carrier watches the reserve grow. At some point the math on settlement changes, and the settlement has nothing to do with what the video actually shows.

This is social inflation with a new accelerant. Jury verdict risk was already climbing. Now the pretrial cost of defending against inflated claims is climbing too, and the tool driving it is the same tool the legal system needs to survive the deepfake era.

Cheap To Allege, Expensive To Disprove In Court

The strategic logic is clean. You do not need to win the authentication fight. You just need to make the other side pay for it. Every hour spent authenticating a real recording or photo is an hour billed, an expert retained, a fee invoice the client sees. Multiply that across numerous digital exhibits in the case and the cost is no longer incidental. It is the case.

And this applies to more than photos, videos, and audio. Medical records, contracts, and receipts are transmitted digitally. Anything can be a fake.

There is currently no meaningful sanction for raising authentication challenges that turn out to be unfounded. Rule 11 sanctions require showing the challenge was frivolous when filed. "We had a good-faith concern about synthetic media" is not frivolous in 2026. It is the ambient anxiety of the entire legal profession. A judge would have to find the challenging party knew the evidence was authentic and challenged it anyway.

Cheap to allege. Expensive to disprove. No downside for being wrong.

The Guardrails Courts And Provenance Standards Could Build

Digital forensics is not going away, and it should not. It is the only answer to a problem the legal system has no other answer for. The question is whether courts will build the procedural infrastructure to keep the right tool from becoming the wrong weapon, whichever side happens to be wielding it.

Courts could require a preliminary showing before triggering full forensic examination. Not proof of forgery. Just a specific, articulable basis for the doubt. A visible anomaly, a metadata inconsistency, a chain-of-custody gap. "Deepfakes exist" would not clear the bar. "This recording has the following characteristics" would.

Cost-shifting would help more. If a party demands forensic authentication and the evidence is conclusively authenticated, that party pays the examination costs. Fee-shifting already exists in discovery contexts. Extending it to authenticity challenges would price out speculative attacks without chilling legitimate ones.

Provenance standards would help most. C2PA, the Coalition for Content Provenance and Authenticity, is an open technical standard founded in 2021 by Adobe, Arm, BBC, Intel, Microsoft and Truepic. It embeds cryptographic signatures into media at the moment of capture. Signed-at-capture media with an unbroken chain of custody is meaningfully harder to challenge in bad faith. The catch is that the signature only travels with the original. The moment a video gets screen-recorded, re-encoded, exported, or forwarded, the provenance data can be stripped, and the derivative copy is back to being just another file someone has to authenticate the hard way. Microsoft's February 2026 Media Integrity and Authentication report reaches the same conclusion: no single provenance method survives every attack, and durable verification has to combine signatures, watermarks and forensic analysis. Much of the evidence in current litigation arrives as derivatives, not originals, and most of it predates the standard entirely. I wrote about this earlier in a February column on what the collapse of visual evidence means for litigation.

Rule 707 cannot take effect before December 1, 2027 even if the Advisory Committee approves it next week. Until they do, the meter runs on whoever has to authenticate the most.

Digital forensics is the answer to deepfakes. It is also, right now, one of the cheapest ways to make the other side spend money. Both things are true at once, and the legal system has to hold them both, or the tool we need will become the lever that settles cases on everything except the merits.

Home

About

Digital Forensics Is the Only Sure Answer to Deepfakes. It's Also the Liar's Dividend's Second Payout