Most worry about deepfakes runs in one direction: someone fabricates a video of you saying something you never said, and people believe it. That risk is real. But two law professors spotted a second, quieter danger years before the technology was good, and it may turn out to be the larger one.
If anything can be faked, then anything can be called fake. The existence of convincing forgeries hands everyone caught on tape a new defence: that was not me, that was AI. Bobby Chesney and Danielle Citron named this in a 2018 paper. They called it the liar’s dividend, and they put the mechanism in one sentence: a skeptical public will be primed to doubt the authenticity of real audio and video evidence.
For a few years this was a prediction. It has since been tried in a courtroom and tested in a laboratory, and both results are worth knowing.
It has already been tried in court
In 2023, the family of a man who died in a Tesla crash sued the company, citing public statements Elon Musk had made about the car’s self-driving capability, including remarks at a 2016 conference that had sat on YouTube for nearly seven years.
Tesla’s lawyers did not say the video was fake. They said it might be. Musk, they wrote in a court filing, like many public figures, is the subject of many “deepfake” videos and audio recordings that purport to show him saying and doing things he never actually said or did.
The judge, Evette Pennypacker, was unimpressed. What Tesla is contending is deeply troubling to the Court, she wrote. Her ruling spelled out the logic of the liar’s dividend better than most academic papers. The company’s position, she wrote, was that because Musk is famous and might be more of a target for deep fakes, his public statements are immune. Public figures could 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.
She ordered Musk to testify under oath, adding that the Court is unwilling to set such a precedent.
Hany Farid, a digital forensics expert at the University of California, Berkeley, summed up the moment: when we entered this age of deepfakes, anybody can deny reality. That is the classic liar’s dividend.
Courts have rules of evidence, cross-examination and the power to compel testimony. Public opinion has none of those.
Does it work on the public? The experiments
Whether crying “fake” actually pays off was tested by political scientists at Purdue and Emory, in a paper published in the American Political Science Review.
They administered five survey experiments to over 15,000 American adults. Participants saw a real video or read a real story about a real politician caught in a real scandal. They were then shown one of several responses attributed to the politician. Some stayed silent. Some apologised. Some claimed the story was misinformation.
The researchers tested two versions of the lie. One invoked informational uncertainty: you cannot trust anything these days, this could easily be a deepfake. The other encouraged oppositional rallying of core supporters: this is a smear by my enemies and the biased media.
Both worked. Claims of misinformation representing both strategies raise politician support across partisan subgroups. It was not only the politician’s own side who were swayed.
And it beat the honest alternatives. These false claims produce greater dividends for politicians than alternative responses to scandal, such as remaining silent or apologizing. The implication is plain: lying bolsters more support than apologising or silence.
The finding that offers some comfort
There was an important limit, and it cuts against the usual fear. The strategies are effective against text-based reports of scandals, but are largely ineffective against video evidence.
When people had read about the scandal, the “fake news” claim moved them. When they had watched the video themselves, it mostly did not. Seeing still counts for something. At least it did when these experiments were run.
A second reassurance: the false claims did not reduce general trust in media. The politician benefited, but the lie did not appear to poison the well for journalism in general.
Why the comfort may be temporary
The video finding deserves a careful reading, because it rests on an assumption that is eroding.
A plausible reading is that video resisted the “it’s a deepfake” defence because participants found it hard to believe the footage was fabricated. That judgement depends on what people believe fakes can do. The more convincing synthetic video they encounter, and the more they hear that detection is impossible, the more plausible the denial becomes.
This is the unpleasant feedback loop. Every warning about deepfakes, including articles like this one, slightly strengthens the hand of anyone who wants to disown genuine footage. Awareness is necessary and it is not free.
What is not known
Whether the video result still holds. The experiments predate the current generation of video generators. Nobody has published a replication showing whether real video remains resistant now that realistic fake video is commonplace.
Whether it works outside politics. The experiments concerned politicians and partisan audiences. Whether an employee, an executive or a public figure in another field gets the same dividend is untested.
How often it is used. There are documented cases of public figures dismissing authentic recordings as AI, but no systematic count, so the real-world scale is unknown.
Whether the dividend grows with repetition, or whether audiences eventually stop believing the excuse.
What follows from this
The liar’s dividend changes what “staying safe from deepfakes” means. It is not only about avoiding being fooled by a fake. It is about not being talked out of something true.
Ask for the basis of the claim. “That’s AI” is an assertion, and it deserves the same scrutiny as the footage. Who says so, and on what evidence? A forensic analysis is evidence. A bare denial from the person who benefits is not.
Look for what a fake would lack. Genuine events tend to leave multiple traces: other cameras, other witnesses, contemporaneous reports, original files with a history. A single clip with no origin is weak whoever is pushing it. A recording corroborated from several directions is strong, however loudly it is denied.
Notice the incentive. The judge’s reasoning applies generally. The person claiming a recording is fake usually has the most to gain from that being true.
If you record things that matter, keep the originals. Provenance (the unedited file, its metadata, where it has been) is what turns a contested clip into evidence. This is the constructive use of the Content Credentials technology: not catching fakes, but letting honest sources prove their material is real.
Do not let scepticism become cynicism. “You can’t trust anything any more” feels sophisticated and is exactly the conclusion the liar needs you to reach. The accurate position is harder: most things are what they appear to be, some are not, and the difference can usually be established with effort.
Common questions
What is the liar’s dividend?
It is the benefit a dishonest person gains from the existence of deepfakes: the ability to dismiss genuine evidence as fake. The term was coined by law professors Bobby Chesney and Danielle Citron in a 2018 paper on the challenges deepfakes pose.
Does claiming something is a deepfake actually work?
In five survey experiments with over 15,000 American adults, politicians who falsely claimed a true scandal was misinformation gained support across partisan subgroups, and gained more than those who apologised or stayed silent.
Does it work against video evidence?
Less well. The experiments found the strategy effective against text-based reports of scandals but largely ineffective against video evidence. Those experiments predate today’s video generators, and the result has not been re-tested.
Has anyone used the deepfake defence in court?
Yes. In a 2023 lawsuit, Tesla’s lawyers argued that recorded statements by Elon Musk might be deepfakes. The judge called the argument deeply troubling, refused to set such a precedent and ordered Musk to testify under oath.
The bottom line
The fakes are only half the threat. The other half is that their existence gives anyone caught on a genuine recording a ready-made excuse, and experiments show the excuse works, outperforming both silence and apology. For now it works better against written reports than against video, and courts have refused to play along. Your defence is to treat “that’s AI” as a claim needing evidence like any other, and to resist the tempting conclusion that nothing can be known.
Did this answer your question?
Thanks. That helps us decide what to fix.
Leave a comment