Cry Deepfake: How “It’s Fake” Became the Perfect Alibi

For years the warning about deepfakes ran in one direction. The fear was that you would watch a video of a politician saying something they never said, believe it, and act on the lie. That fear was reasonable, and the fakes kept getting better. But it turns out we were bracing for the wrong hit. The danger is no longer that a fake video fools you. It is that a real one no longer convinces anyone. The person caught on camera doing something ugly now has a brand new defence, and it costs them nothing. They just say it is AI. They do not need to prove it. They only need you to hesitate.

There is a name for it. Legal scholars Bobby Chesney and Danielle Citron called it the "liar's dividend": the payout a dishonest actor collects simply because deepfakes exist at all. They do not have to make one. They just have to point at the possibility. The work is already done by everyone else's fakes. The liar just shows up and cashes the cheque.

I wrote a while back about how the Kremlin spent twenty years trying to kill the very idea of objective truth, not by selling one big lie but by drowning people in so much noise that they gave up telling true from false. The liar's dividend is what happens when that twenty-year project gets handed to every guilty politician, every cornered executive, and every regime on earth, for free, as a built-in feature of the technology. You no longer need a troll farm. You need one sentence.

Strip away the jargon and the engine is just economics. Saying "that could be a deepfake" is free. No evidence, no expertise, no money, no time. Anyone can say it about anything, instantly. Proving that a piece of footage is genuine is a different universe of effort: forensic analysis, chain-of-custody records, expert testimony, and even then you get a probability, not a certainty. The burden of proof flips. It is no longer on the liar to show the footage is fake. It is on you to show it is real.

So the two sides are not playing the same game. The denial is cheap and the rebuttal is costly, and in a feed that scrolls past in seconds, cheap wins. Ordinary disinformation manufactures false content; the liar's dividend does the opposite, poisoning trust in true content. Everything genuine now carries an asterisk.

To see how this plays out, it helps to stop talking in theory. Over the past year three very different operations showed three different ways to break the truth. One built fake people. One drowned a war in noise. One turned the doubt itself into a doctrine.

In May 2026, Tommy Robinson held his second "Unite the Kingdom" rally in central London. At one point, a track played to the crowd and Robinson grinned and told them, "Even the AI characters are on our side in this battle." The artist he was talking about is called Danny Bones, and Danny Bones is not a person. He is an AI-generated rapper.

He was built by an outfit called the Node Project, which the Bureau of Investigative Journalism revealed had been commissioned by Advance UK, the party launched by former Reform UK co-deputy leader Ben Habib. The character was paid for to produce campaign material before a by-election. Online, Danny Bones runs a verified account, posts anti-immigration content that splices historical war footage with images of British nationalists, and cheerfully tags and boosts real political figures like Robinson and Rupert Lowe. He has fans. He has a persona. He has a worldview. What he does not have is a body, a conscience, or anyone who has to answer for a word he says.

That is the first way to break the truth. You do not fake a real person doing something. You build a fake person from scratch and let them do the work a real activist would do, except this one never tires, never goes off message, and can never be held to account, because there is no one there.

There were two other moves at that same rally worth noting. Supporters shared an AI-generated image of an enormous crowd that the real, far smaller attendance flatly contradicted. And when unflattering clips circulated, at least one figure linked to the rally claimed real footage of him was itself AI-generated, posting "real versus fake" comparisons to muddy it. Manufacture a bigger crowd, launder the message through a synthetic rapper, deny the inconvenient real clips. One movement, one Saturday, ran all three.

Here is the obvious objection. Danny Bones is labelled. YouTube slaps a transparency tag on the videos, Spotify says the tracks passed human review, TikTok pulled one of the accounts. So where is the harm? People know it is AI. The disclosure is right there.

That is exactly the problem, and it is worth sitting with, because it quietly demolishes the fix that nearly every government is currently betting on.

Start with the simple version. A label and a person's attention do not occupy the same space. The tag sits in a corner, greyed out, while the persona, the music, the anger, and the message go straight in. In a scrolling feed, a labelled AI artist functions as a real one, because nobody stops to read the label. Disclosure assumes the audience pauses, notices, and updates. They do not. They keep scrolling.

But it gets worse, and this is the part that should genuinely worry the people writing the rules. When labels are noticed, the research says they often backfire. A 2026 study in the Journal of Science Communication found what the authors called a "truth-falsity crossover effect": an AI-generated label lowered how much people trusted true information while actually nudging up how credible they found false claims. People who already distrusted AI punished the true, labelled material the hardest. The label did not teach people to sort true from false. It taught them to distrust whatever wore the tag and relax around whatever did not. It does not create careful readers. It creates blanket cynics, and blanket cynicism is the precise soil the liar's dividend grows in.

Now look at where policy is heading. The EU's AI Act is moving to mandate exactly this kind of labelling, with a Code of Practice due in mid-2026. Meta has rolled out an "AI Info" tag, TikTok has its creator guidelines, and the whole regulatory bet is that disclosure will inoculate the public. The evidence suggests the cure may feed the disease. It is the same trick I wrote about with online safety laws, a measure sold as protection that quietly hands ground to the thing it claims to fight. The label lets the platform, the party, and the funder all claim good faith while the synthetic messenger does its job unbothered. Compliance, not consent.

The second way to break the truth showed up in a shooting war.

When the United States and Israel launched Operation Epic Fury against Iran on 28 February 2026, killing the Supreme Leader and pounding Iran's nuclear and military sites, Iran was losing badly on the battlefield. Its conventional forces had been ground down over months. And yet within weeks, disinformation researchers were writing, with some alarm, that Iran was winning. Not the war. The vibe war.

That is the term that stuck, because the campaign did not look like classic propaganda at all. It was AI-generated Lego rap videos, embassy accounts posting like shitposters, and culturally fluent English-language memes aimed straight at the Trump administration. The Atlantic Council's Digital Forensic Research Lab made the key observation: a lot of this material was not even trying to fool anyone. It was obviously fake, and that was fine, because the goal was not deception. The goal was to set a mood, to make Iran feel like the scrappy underdog landing punches on a bloated superpower. This is a step beyond the firehose of falsehood in the Kremlin's playbook. It is propaganda that openly admits it is fake and works anyway.

Alongside the vibes ran a genuine flood of fakes. The New York Times identified more than 110 unique pro-Iran deepfakes in a two-week span, much of it produced by Iran-linked networks and amplified through Russian and Chinese ecosystems. One fabricated clip of an Iranian jet facing down a US warship, later identified as video game footage, racked up more than seven million views. The fakes arrived faster than anyone could debunk them.

And here is where the war ties straight back to our theme. The flood was not only offence. Chatham House noted that Iran's internet shutdown made it easy for the regime to wave away real content as Western-manufactured deepfakes. Once the information space is thick enough with synthetic junk, a regime can dismiss authentic documentation of its own abuses as just more AI. The fakes built the fog, and the fog became a shield. That is the liar's dividend operating at the scale of a state at war.

The third way is the most deliberate, and it comes, predictably, from Moscow.

Russia's deepfake habit is not new. Back in March 2022, just after the full-scale invasion, pro-Russian actors hacked a Ukrainian news site and pushed out a crude deepfake of President Zelenskyy telling his soldiers to surrender. It was clumsy and got debunked fast. The interesting thing is how far the operation has travelled since. By late 2025, deepfakes made with OpenAI's Sora were showing Ukrainian soldiers weeping and surrendering en masse, and these ones looked real to the naked eye, with few of the old telltale glitches. In August 2025, a network tracked as Storm-1679 was impersonating ABC, the BBC, POLITICO and Netflix with AI-cloned voices, convincing enough that prominent people shared the fabrications to millions.

But the part that matters most is the stated goal. In April 2026, Ukraine's Center for Countering Disinformation, working with the detection firm Sensity AI, described a structured Russian system of more than a thousand synthetic videos. They called it a "narrative kill chain," a modular machine pumping out tailored fakes for different audiences: clips to demoralise Ukrainian soldiers, clips to exhaust civilians, clips to make Western voters sick of funding the war.

And the assessment spelled out the real aim in words that could be the thesis of this entire piece. Russia's ultimate goal, they wrote, is not to make you believe any particular message. It is to create so much information chaos that any truth can be dismissed as a deepfake, which lets the aggressor escape accountability for real crimes by casting doubt on any evidence at all.

Read that again, because it is the whole thing stated plainly by the people on the receiving end. The fakes are not the point. The doubt is the point. Russia is not building a thousand videos to win a thousand arguments. It is building them to make argument itself feel pointless, so that when real evidence of a real war crime appears, the reflex is already installed: probably fake, who knows, who can say. That is not a side effect. It is the doctrine.

If you want to see how this scales down from geopolitics to a single human life, look at the one place built specifically to decide what is real: a court of law. It has a referee, formal rules of evidence, and serious stakes. And it is already buckling.

In a wrongful-death case against Tesla over its Autopilot system, lawyers balked at admitting real recorded statements from Elon Musk about the feature's safety, floating the idea that the video could be a deepfake. The court was not impressed and warned of the obvious slippery slope: if every famous person can hide behind "it might be a deepfake" to disown things they actually said and did, recorded evidence stops meaning anything. The denial did not have to succeed to be worth trying. It just had to introduce a wobble.

In the prosecution of a January 6 defendant, the defence raised the possibility that footage had been AI-manipulated, despite having nothing to support the claim. That is the dividend in its purest courtroom form. You do not bring proof. You bring doubt, for free, and dare the other side to spend a fortune dispelling it.

Then there is the mirror image, which is just as unsettling. In a California housing dispute, Mendones v. Cushman & Wakefield, a judge realised that a video submitted as genuine witness testimony was itself an AI deepfake. The face barely moved, the expressions repeated, the cadence was off. She caught it and threw the case out. Good. But notice why she caught it: because the fake was still crude enough to show seams. That defence worked exactly once, on a bad fake. The next generation will not twitch and repeat. And when the fakes leave no seams, the only thing left is the doubt, applied to everything, real and fake alike.

The instinct at this point is to reach for a technical fix. If fakes are the problem, build software that spots fakes. Train a really good detector, run everything through it, done. I understand the appeal, and I am telling you it does not work, for reasons baked into how this technology is built.

The systems that generate convincing fakes improve by being tested against systems that try to catch them. That is close to the original idea behind the technology: one network generates, another judges, and the generator gets better and better at slipping past the judge. So every detector you build becomes, in effect, free training material for the next generation of fakes. You are not ending the arms race. You are sponsoring the other side.

The economics are lopsided in the same direction as everything else here. Generating fakes is cheap and scales to infinity. Verifying content forensically is slow, costs real money, and does not scale to the volume of an internet where a single war produces a hundred new fakes a week. And even when detection works, it hands you a probability, not a verdict. "Likely synthetic, 80 percent confidence" is not something a court can convict on or a newsroom can stake its reputation on. Studies keep finding that ordinary people spot fakes at rates barely better than a coin flip.

So detection treats this as a technical problem. It is not a technical problem. It is a problem about trust, about how humans decide what is real together, and you cannot patch that with a classifier.

There is a comforting story we tell about disinformation: that people believe false things because they have not seen the facts yet. Show them the evidence, correct the record, and they will come around. Decades of research say this is mostly wrong, and the way it is wrong explains why "it's fake" works so well.

When a fact collides with something central to who a person is, the fact usually loses. Psychologists call this motivated reasoning, and the rough idea is that people do not weigh evidence like a judge. They argue like a lawyer who already has a client. The conclusion comes first, and the reasoning works backwards to defend it. Faced with proof they would rather not accept, people go looking for a reason to throw it out, and they almost always find one. The source is biased. The study was flawed. The video was edited. It's a deepfake.

Here is the part that should bother you, because it bothered the researchers. You might assume this is a problem of ignorance that better education would fix. The opposite seems closer to the truth. In a well-known set of studies, the people best equipped to reason through evidence were not the ones most likely to reach the right answer on politically loaded questions. They were the ones who polarised the hardest. Give a skilled reasoner a fact they do not want, and their skill becomes a tool for explaining it away more cleverly than anyone else could. Intelligence does not pull people toward the truth on these questions. It gives them better lawyers for the conclusion they already wanted.

I should be honest about the science, because this is exactly the kind of tidy psychological story that gets oversold. The most dramatic version, the so-called backfire effect, the claim that correcting people makes them believe the falsehood even more strongly, has largely failed to hold up when researchers tried to reproduce it. It happens, but it is rarer than the headlines suggested, and even the sharper findings about clever people polarising harder have their critics. So take the strong claims with caution. But the mild version is more than enough here, and it is not seriously in dispute: people reason toward the conclusions their identity wants, and contrary evidence is something to be defended against, not absorbed.

Sit that next to the liar's dividend and the whole thing clicks into place. "It's fake" was never required to win an argument. It does not have to be persuasive, or evidenced, or even plausible. It only has to hand people who already agree a sentence they can repeat. The work of believing was done long before the denial showed up. The supporter who does not want the footage to be real is not looking to be convinced. They are looking for permission, and three words grant it. That is why debunking the denial changes so little. You are aiming facts at a person who was never deciding on the facts.

Step back and notice who collects the dividend and who covers the cost, because it is the same lopsided story I keep coming back to on this blog. The powerful win and everyone else pays.

Denial is a luxury good. A politician, a celebrity, a corporation with a legal team, these are the people for whom "it must be a deepfake" is plausible, because they are exactly the kind of high-value target who realistically would be faked. Fame becomes a shield, just as the Tesla court warned. The more famous and powerful you are, the more credible your denial, which is precisely backwards from how accountability is supposed to work.

And the people who lose are the ones with no leverage. The abuse survivor whose real evidence gets waved off as AI. The whistleblower whose genuine recording is dismissed. The journalist whose verified footage is drowned under a coordinated "that's fake" campaign. The ordinary litigant who cannot afford a forensic expert to prove their own video is real.

The civic cost is the biggest one and the hardest to see. When nothing can be proven, accountability becomes optional, and optional accountability is just impunity with better PR. That is the destination all three of our cases are heading toward. The UK case manufactures consent and mislabels it. The Iran case floods the zone and hides behind the flood. The Russia case industrialises the doubt on purpose. Different methods, one outcome: a world where the question "is this real?" stops having an answer anyone trusts.

I do not want to end on a shrug, so here is the honest version of what helps, including where it falls short.

The most promising shift is from chasing fakes to proving real things. Instead of asking software "does this look fake," you attach a secure, cryptographic record of origin to genuine content at the moment it is captured, so the question becomes "can you show me the chain of custody." Provenance standards like C2PA try to do exactly this, and they are a better bet than detection because they work with the asymmetry instead of against it. But the honest caveat is that everything we just learned about labels applies here too. Provenance only matters if people check it and update their trust, and the evidence on whether they do is not encouraging. It is necessary, and nowhere near sufficient.

The courts can help by raising the price of a baseless "it's AI" claim. Judge Paul Grimm and others have proposed clearer rules that separate honestly disclosed AI material from a litigant simply crying deepfake to dodge real evidence. Courts already sanction people for submitting fabricated evidence. Crying fake without any basis should carry a similar cost, because right now the tactic is free, and free is the whole problem.

But the deepest fix is neither technical nor legal, and it follows from both the label research and the psychology. The reflex the liar's dividend feeds on is the easy, lazy version of skepticism, the shrug that treats everything as equally suspect. Real skepticism is harder and rarer: the willingness to follow evidence somewhere you would rather it did not go. That is a problem of culture and habit, not of metadata.

Which brings it back to where the propaganda piece ended. The antidote was never going to be a cleverer detector or a louder warning label. It is a public that still believes shared reality is worth the effort of defending, and institutions willing to make denial cost something again. The fakes are not really the threat. The threat is the day we stop believing that the difference between true and false is worth fighting for. Three operations, on three fronts, are betting we have already given up. The whole point is to prove them wrong.