When Seeing Is No Longer Believing
The liar’s dividend and why journalists need to show how they verify video in the age of AI
By Yasmin Ali
When Real Footage Becomes Questionable
In March 2026, during the war with Iran, rumors spread online that Israeli Prime Minister Benjamin Netanyahu had been killed or seriously injured. When his office released video of a March 12 press briefing, people on social media said the footage was AI-generated, pointing to one frame where he seemed to have six fingers.
I saw the Netanyahu clips myself on TikTok, and honestly, they worried me. The way it was animated and the way it looked made me think it was fake. When I saw the video, I immediately knew it was AI, because there’s no angle that could possibly show him having six fingers except that. I still don’t fully believe it.
But PolitiFact looked at the video and found that the “sixth finger” was just a trick of the lighting, and several news outlets covered that same briefing live. Netanyahu later posted another video of himself buying coffee and holding up five fingers, and people said that one was fake too.
What Is the ‘Liar’s Dividend’?
The idea comes from legal scholars Robert Chesney and Danielle Citron, who wrote about it in 2019 in “Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security”. They called it the “liar’s dividend”: once people know video and audio can be faked, someone caught on a real recording can just say it’s a deepfake. That claim can create enough doubt to get them off the hook, even when the recording is real.
To me, this seems scary because it means people can get away with doing bad things. It’s harder to tell whether or not a person is lying. It makes liars seem more truthful, and you can’t really tell who’s honest anymore.
For example, when inappropriate pictures of celebrities get leaked, a lot of them will say “that’s AI.” Sometimes that’s actually true, because AI can make fake pictures of real people. But the problem is that you can’t tell whether the person is telling the truth or lying. The same thing could happen if someone got caught on video saying something offensive and just said “that’s AI” to avoid being held accountable.
And people already don’t trust AI in the news. The Reuters Institute’s 2024 Digital News Report, in a chapter by Amy Ross Arguedas, found that only 36 percent of people across 28 countries are actually comfortable with news made by journalists using AI help, and only 19 percent with news made mostly by AI.
Honestly, I wouldn’t want a news source to use AI, because I feel like only humans can really express what happens around them. Compared to AI, they know what happens better and they’re able to put that into words better.
Research Shows That Denial Works
Researchers have actually tested this. For their study “The Liar’s Dividend: Can Politicians Claim Misinformation to Evade Accountability?” in the “American Political Science Review”, Kaylyn Jackson Schiff, Daniel S. Schiff and Natália S. Bueno surveyed more than 15,000 U.S. adults in five experiments.
When a politician caught in a real scandal said it was misinformation, people supported them even more for it, including people from the other political party. So it worked better than apologizing. I think denying has always worked better than apologizing, because it makes the person come across as stronger. In the 2021 experiments, the denial mostly didn’t work when the scandal was caught on video, but in a 2022 follow-up, one version of it did. The authors pointed out that awareness of deepfakes had grown a lot between the two studies.
In a 2026 study in Psychology of Popular Media, Grohmann, Halle and Appel showed people a real video of a politician. The politician falsely said it was a deepfake. Viewers started believing that the real video was fake, and they rated the politician as a better leader.
The good news is that in the Schiff study, when a fact-check was attached to the politician’s claim, the trick stopped working. I think fact-checking works because it proves the politician was lying.
Breaking News Is Where It Bites
It goes the other way too. In August 2025, NewsNation host Chris Cuomo reposted a fake video on X that seemed to show Rep. Alexandria Ocasio-Cortez criticizing an ad with actor Sydney Sweeney. The clip was labeled as an AI parody, but Cuomo responded like it was real. After she called it a deepfake, he deleted his post and admitted the mistake, as The Guardian and The Independent reported on Aug. 7, 2025.
I think it’s crazy that even journalists are falling for it now and not realizing how many clips are AI. When journalists fall for fakes, people start to rely less on journalists, because it’s not the same as before. If I were the journalist, I would have taken a better look at it and checked my sources before taking any big risks, to see whether or not it’s real or AI.
Why Detection Tools Cannot Be the Final Answer
AI-detection tools can help, but they make mistakes in both directions. They can flag a real video as fake, which is called a false positive, or miss a video that really is fake, which is called a false negative.
A 2024 study by Saniat Javid Sohrawardi, Y. Kelly Wu, Andrea Hickerson and Matthew Wright, “Dungeons & Deepfakes,” gave journalists a breaking-news scenario and an AI detection tool that was sometimes wrong. The journalists used normal methods first, like contacting sources, but when they were under time pressure they leaned on the tool more. The researchers warned that this raises the risk of publishing fake content as real.
I think people should take more time and effort to verify whether a source is real or not, and not rush. That way it could stop AI fakes from spreading and increasing. Checking a clip means finding the earliest version of it, contacting whoever recorded it, confirming the date and location, and comparing it with other footage.
Expert Perspective: Hany Farid
Hany Farid is a professor of computer science at Dartmouth College who studies digital forensics. He taught at UC Berkeley’s School of Information until June 2026, when he returned to Dartmouth. He talked about deepfakes in a Feb. 17, 2026, interview with Scientific American and in a Berkeley Talks episode published by UC Berkeley News on April 6, 2026.
In the Scientific American interview, Farid said people are only slightly better than chance at telling AI-made faces, voices and videos from real ones, and they’re often confident even when they’re wrong. I think it’s very obvious for Gen Z to know whether or not a video is AI compared to middle-aged and older adults. His research says otherwise, but I still think younger people have an advantage, because my friends and I know more about what AI looks like. My parents usually share AI videos, and they don’t realize it’s AI until I point it out to them.
I agree that AI will continue to improve, as it’s already harder to differentiate between AI and real videos than it was before.
Get Yasmin Ali’s stories in your inbox
Join Medium for free to get updates from this writer.
In the UC Berkeley News talk, his advice to regular people was simple: “Stop getting your news from social media.” I agree, because sometimes sources gather information incorrectly or try to deceive people into thinking what they want them to think. I feel like it’s better to do research and read actual news articles about it instead of believing everything you see online.
But honestly, it’s not really realistic, because my age really depends on social media. There’s no other major source where we can get our news and learn more about what’s going on in the world. Even TikTok, for example, I feel like it hides a lot of valuable news, because it’s able to filter it and only show people what they want them to see.
In the same UC Berkeley News talk, Farid also argued that solutions should focus on the systems that profit from harmful content, including the platforms and ad networks that help it spread. I think TikTok should be doing more, because they let AI run freely across their platform, and it’s slowly taking over, especially AI stories.
People tend to look at them more and are entertained by them, even though they’re not natural and not real. People are slowly getting more used to it and enjoying it, even me. I read an AI story, and I enjoyed it and kept reading it and wanted more. But I feel like that’s also wrong, because it’s not natural and it’s not human-made.
Provenance and Showing the Work
One tool that helps is provenance, which is basically a tracking history for a photo or video. The Coalition for Content Provenance and Authenticity (C2PA) made a system called Content Credentials that can attach information to a file about what camera took it, when, and how it was edited.
It’s not perfect, though. A real video can lose that history when it gets reposted on social media, and a video with a clean history can still be shared with the wrong caption or out of context.
I don’t think people are gonna care whether or not a video is labeled as edited. I feel like it’s more about what they see with their own eyes. People like to judge it themselves and decide whether it’s fake or real, or whether it’s a reliable news source or reliable information.
Newsrooms should also show readers how they checked a video. I would trust a story more if the source is reliable and there is proof that the video is recent, and not just a repeat of something that happened two or three years ago. The Schiff study backs this up: when a fact-check was attached to a misinformation claim, the liar’s dividend went away.
What Newsrooms Should Do Next
My advice to journalists is to take more time when identifying whether something is fake or real. Rushing really does confuse things, and if you find out later that your source was AI, it makes it harder to publish things and draw attention, because no one will trust your source. That’s what happened to Chris Cuomo, and it’s what the Dungeons & Deepfakes study warned about.
My advice to regular people scrolling on TikTok is to go on the profile and check if the other videos seem AI. That way you’ll know if the video you watched is inorganic or unoriginal. And like Farid said in his UC Berkeley News talk, social media shouldn’t be the only place you get your news.
Conclusion
The main thing I learned from this paper is to stop believing everything I see right away, including news sources, and to do my own research and dive deeper into things to find out whether something is recent and whether it’s true.
What worries me most is the news, because the entire world tends to look at the news, especially in areas that have less access to the internet and rely more on television. That’s their reliable source. So it’s a big worry that people can fabricate videos and make it look like something bad is happening when nothing is going on, just to capture people’s attention.
Even now, I still don’t fully trust the Netanyahu video. That’s what the liar’s dividend does: it makes it hard to tell who’s honest anymore. That’s why journalists need to slow down and show how they checked their work.
Sources
Chesney, R., & Citron, D. K. (2019). Deep fakes: A looming challenge for privacy, democracy, and national security. California Law Review, 107(6), 1753–1820. https://doi.org/10.15779/Z38RV0D15J
Coalition for Content Provenance and Authenticity. (n.d.). C2PA specifications and Content Credentials. https://spec.c2pa.org/specifications/
Grohmann, R., Halle, L., & Appel, M. (2026). Deepfake! A liar’s dividend for audiovisual material. Psychology of Popular Media. Advance online publication. https://doi.org/10.1037/ppm0000665
PolitiFact. (2026, March 13). No, Netanyahu didn’t have six fingers during his press conference. https://www.politifact.com/factchecks/2026/mar/13/social-media/netanyahu-dead-video-artificial-intelligence/
Ross Arguedas, A. (2024, June 17). Public attitudes towards the use of AI in journalism. Reuters Institute for the Study of Journalism, University of Oxford. https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2024/public-attitudes-towards-use-ai-and-journalism
Schiff, K. J., Schiff, D. S., & Bueno, N. S. (2025). The liar’s dividend: Can politicians claim misinformation to evade accountability? American Political Science Review, 119(1), 71–90. https://doi.org/10.1017/S0003055423001454
Scientific American. (2026, February 17). A deepfake can ruin you before breakfast [Interview with Hany Farid]. https://www.scientificamerican.com/article/deepfakes-are-getting-faster-than-fact-checks-says-digital-forensics-expert/
Sohrawardi, S. J., Wu, Y. K., Hickerson, A., & Wright, M. (2024). Dungeons & deepfakes: Using scenario-based role-play to study journalists’ behavior towards using AI-based verification tools for video content. Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3613904.3641973
Sommerlad, J. (2025, August 7). Chris Cuomo trolled for falling for AOC deepfake video about Sydney Sweeney. The Independent. https://www.independent.co.uk/news/world/americas/us-politics/chris-cuomo-aoc-sydney-sweeney-jeans-b2803523.html
The Guardian. (2025, August 7). Chris Cuomo mocked the response after falling for deepfake AOC video. https://www.theguardian.com/us-news/2025/aug/07/chris-cuomo-alexandria-ocasio-cortez-deepfake
UC Berkeley News. (2026, April 6). Berkeley Talks: The erosion of shared reality in the age of deepfakes. https://news.berkeley.edu/2026/04/06/berkeley-talks-hany-farid-on-deepfakes/
UC Berkeley School of Information. (2026, March 3). Professor Hany Farid to depart UC Berkeley. https://www.ischool.berkeley.edu/news/2026/professor-hany-farid-depart-uc-berkeley
