What Is the Liar’s Dividend? Origins, Examples, and Response

The liar’s dividend is the benefit a public figure gets by dismissing real, incriminating audio or video as a deepfake. As more people learn that convincing synthetic media exists, a flat denial — “that’s AI,” “that’s fake” — becomes more plausible, and the person caught on tape can shift the conversation from what they did to whether the recording is real. Law professors Robert Chesney and Danielle Citron named the dynamic in their 2019 California Law Review article on deepfakes, and researchers, courts, and election officials have been tracking it ever since.¹

Where the Term Comes From

Chesney and Citron introduced the concept while cataloguing the harms of deepfake technology: fraud, nonconsensual pornography, election distortion, national-security threats. The liar’s dividend sat in its own category as an indirect harm. No fake has to actually be produced for the dividend to exist. The widespread knowledge that fakes are possible is enough to hand anyone a ready-made excuse when authentic evidence surfaces.² The dynamic feeds itself. Every real deepfake that circulates makes the next denial a little more believable.

The two scholars expanded the argument in a Foreign Affairs piece framing deepfakes as the arrival of “post-truth geopolitics,” and that pair of publications is where most later writing on the subject starts.³

How It Actually Works

A 2024 study in the American Political Science Review by researchers at Purdue’s Governance and Responsible AI Lab put the idea to an empirical test. Across five experiments with more than 15,000 American adults, participants were shown real political scandals and the politician’s response. Denials that labeled the scandal “fake news” or a “deepfake” produced more public support than apologies or silence.⁴

The researchers identified two mechanisms doing the work:

Informational uncertainty. The politician casts doubt on the entire information environment, suggesting any recording could be fabricated, so voters can’t be sure what’s true.
Oppositional rallying. The politician reframes the scandal as a partisan attack by hostile media or opponents, which triggers loyalty among supporters.

Format matters. The dividend was strong when scandals were reported in text, with denials cutting opposition by 10 to 12 percentage points. It was much weaker when participants actually watched video of the scandal.⁵ As public awareness of video deepfakes grows, the researchers warned, even that cushion could shrink.

One finding complicates the usual story. Crying “fake news” boosted the politician’s standing, but it did not consistently drive down general trust in the media or convince voters the scandal never happened. The dividend works more as a tribal rallying mechanism than as a truth-killer. Supporters close ranks without necessarily believing the denial.⁴

Real Examples of People Using It

Politicians, corporations, and defendants have already reached for the dividend in high-profile situations.

Political Denials

After the 2016 release of the Access Hollywood tape, Donald Trump reportedly suggested in 2017 that the recording was not authentic.⁶ Michigan mayor Jim Fouts labeled audio tapes of him making derogatory remarks about women and Black people “phony, engineered tapes,” though experts confirmed the recordings were genuine. In Spain, former Foreign Minister Alfonso Dastis dismissed photographs of police violence during the Catalonia crisis as “fake photos.”

In Malaysia in 2019, Economic Affairs Minister Azmin Ali denied being the man in a widely circulated sex video, calling it a “nefarious plot.” Prime Minister Mahathir Mohamad reinforced the denial: “I do not believe this is genuine. Nowadays, we can do a lot of things with videos and pictures if we are good at it.”⁷ No deepfake was ever proven to exist, but the mere availability of the excuse provided cover.⁸

The Gabon Episode

A sharper illustration played out in Gabon. President Ali Bongo had been out of public view since suffering a stroke in October 2018. When the government released a New Year’s video of him, critics seized on his stiff appearance and strange speech, and politician Bruno Ben Moubamba publicly alleged the footage was a deepfake. A week later, the military attempted a coup, citing the video’s oddness as evidence the president was unfit.⁹ Forensic analysts at Deeptrace Labs later concluded the video was likely authentic, with Bongo’s appearance attributable to his stroke.¹⁰

In Court

Lawyers have begun invoking the dividend as a litigation tactic. During the prosecution of Guy Reffitt for his role in the January 6, 2021 Capitol breach, his defense attorney asked an FBI agent whether video of Reffitt carrying a handgun could have been AI-generated, while conceding there was no factual basis for the suggestion. Reffitt was sentenced to 87 months in prison.⁶

In Huang v. Tesla, a wrongful-death case in Santa Clara County Superior Court, Tesla’s lawyers argued that recordings of Elon Musk discussing Autopilot safety could not be authenticated because deepfakes exist. Judge Evette Pennypacker rejected that position in blunt terms, writing that it would allow Musk and others to “simply say whatever they like in the public domain, then hide behind the potential for their recorded statements being a deep fake to avoid taking ownership of what they did actually say and do.”¹¹¹² That ruling is one of the first judicial pushbacks on the tactic.

The opposite problem has also surfaced. In Mendones v. Cushman & Wakefield in Alameda County, self-represented plaintiffs submitted a video as witness testimony that turned out to be a deepfake. Judge Victoria Kolakowski noticed the witness’s near-motionless face, and forensic examination showed the file’s metadata claimed capture on an iPhone 6, a device too old to have produced it. The case was dismissed on September 9, 2025.¹³ Both ends of the problem — real evidence dismissed as fake, and fake evidence passed off as real — are now live issues for judges.

Why the Public Falls for It

A 2026 systematic review in Frontiers in Political Science, synthesizing 21 studies, called the broader effect a “skepticism tax” — a cognitive burden on everyone consuming media in an environment where any recording might be synthetic. Across those studies, 86% found a significant positive relationship between deepfake exposure and epistemic uncertainty, with video producing the largest effects, then audio, then images.¹⁴

Yale researchers John Ternovski, Joshua Kalla, and P. M. Aronow found that simply informing people about deepfakes did not improve their ability to spot fakes, but it did make them more likely to dismiss real videos as fabricated.¹⁵ An August 2023 YouGov survey put 85% of Americans in the “very” or “somewhat” concerned camp about misleading deepfakes.⁶ That concern is part of what makes the dividend work. A public primed to worry about fakes is a public primed to accept denials. The Frontiers review also found that political moderates bear the brunt of the skepticism tax, while highly partisan viewers hold stable beliefs regardless of exposure, which lines up with why oppositional rallying succeeds.¹⁴

What’s Being Done About It

Responses are moving on three tracks: evidence rules, authentication technology, and public norms.

On the rules side, the U.S. Judicial Conference’s Advisory Committee on Evidence Rules proposed a new Federal Rule of Evidence 707 in November 2024, governing “machine-generated evidence.” It would require anyone offering AI-generated evidence to show it was based on sufficient data and produced through reliable methods, with a public comment period running through February 16, 2026.¹⁶ Critics point out that Rule 707 only reaches evidence a party admits is AI-generated, which leaves the harder problem untouched: evidence someone falsely claims is a deepfake. Loyola Law School professor Rebecca Delfino has proposed a separate amendment to Rule 901 that would shift the burden to the proponent to prove authenticity once an opponent raises a credible deepfake challenge, with the judge rather than the jury deciding.¹⁷

On the technology side, the Coalition for Content Provenance and Authenticity, or C2PA, is building a standard for cryptographically signed metadata attached to media at the point of creation. The coalition describes it as a “nutrition label” for digital content: it records who made the file, when, and with what tools, and the cryptographic seal breaks if the file is altered.¹⁸ Camera makers including Leica, Nikon, and Canon are integrating it into hardware. A stronger version called Durable Content Credentials adds watermarking and fingerprint matching so provenance survives when metadata is stripped by social platforms.¹⁹ The EU AI Act will require AI-generated content to be labeled by August 2026.²⁰ Adoption is currently opt-in, and AI detectors are locked in an arms race with generators, so detection alone is unreliable.⁶

On the norms side, the Brennan Center has urged political parties to pledge to withdraw support from candidates who falsely label real content as AI-generated, and has pushed digital literacy programs that teach “truth discernment” rather than blanket skepticism.⁶ The distinction matters. Research keeps finding that telling people deepfakes exist mostly makes them suspicious of everything, authentic footage included, without improving their ability to identify fakes.¹⁵ Overhyping the threat can itself enlarge the dividend it is meant to shrink.

One narrower federal law is now on the books: the Take It Down Act, signed May 19, 2025, criminalizes publication of nonconsensual intimate imagery, including AI-generated “digital forgeries,” and requires platforms to remove flagged content within 48 hours.²¹ It addresses one harm from deepfakes but does nothing to establish general authentication standards or constrain the political use of the liar’s dividend.

Northeastern University professor Don Fallis has observed that in an environment saturated with synthetic media, it is “epistemically irresponsible” to automatically trust that what a video shows actually happened — a reasonable intellectual stance that nonetheless hands a weapon to anyone who wants real evidence dismissed.⁶ That paradox is the heart of the liar’s dividend, and it is why the responses that seem to work best are the ones that make authenticity provable, rather than the ones that just warn people fakes exist.

1
California Law Review. Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security
2
Boston University School of Law. Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security
3
JSTOR. Deepfakes and the New Disinformation War
4
Cambridge University Press. The Liar’s Dividend: Can Politicians Claim Misinformation to Evade Accountability?
5
Brookings Institution. Watch Out for False Claims of Deepfakes and Actual Deepfakes This Election Year
6
Brennan Center for Justice. Deepfakes, Elections, and Shrinking the Liar’s Dividend
7
The Straits Times. Malaysia’s Minister Azmin Denies He Is Man in Gay Sex Video
8
The Star (Malaysia). Is It Azmin or a Deepfake?
9
Mother Jones. Deepfake Gabon Ali Bongo
10
MIT Technology Review. The Biggest Threat of Deepfakes Isn’t the Deepfakes Themselves
11
Ars Technica (Court Document). Huang v. Tesla Deepfake Ruling
12
The Guardian. Elon Musk’s Statements Could Be Deepfakes, Tesla Defence Lawyers Tell Court
13
NBC News. AI-Generated Evidence: Deepfake Use in Law as Judges Object
14
Frontiers in Political Science. Skepticism Tax and Epistemic Uncertainty
15
Yale Institution for Social and Policy Studies. The Negative Consequences of Informing Voters About Deepfakes
16
Quinn Emanuel. Adapting the Rules of Evidence for the Age of AI
17
U.S. Courts. Suggestion from Prof. Rebecca Delfino – Rule 901
18
Content Authenticity Initiative. How It Works
19
U.S. Department of Defense. Content Credentials Cybersecurity Information Sheet
20
Australian Cyber Security Centre. Content Credentials: Strengthening Multimedia Integrity in the Generative AI Era
21
National Association of Attorneys General. Congress’s Attempt to Criminalize Nonconsensual Intimate Imagery