Synthetic media and the authenticity crisis
Generative tools broke something in both directions at once. Convincing fakes now circulate during live news events and reach millions before any correction lands, and at the same time authentic photographs and recordings can be waved away as AI by anyone they inconvenience. These files cover both halves: specific fakes and how each was caught, and the newer move of calling the genuine record synthetic. Detection is part of the story too, including watermarking's first public catch in a live news cycle.
Reference: Wikipedia
AI companies are quietly buying up millions of physical books, including rare and obscure titles, slicing them apart to scan them for training data and destroying them, under a secret programme deliberately kept from the public
In July 2026 a claim spread across Instagram, Facebook and Reddit that AI companies were buying obscure and rare physical books in bulk, cutting them apart to scan them, and destroying them to build chatbots, under an internal programme whose own documentation said the work should be kept quiet. Unusually for something that travels that well, it is largely true, and the evidence is not a leak but a court file. During a copyright lawsuit brought by authors in 2024, a judge ordered Anthropic to unseal internal records. Those records, public since January 2026, describe Project Panama, defined in an April 2024 memo as the effort to destructively scan all the books in the world, with a stated reason for the codename: the company did not want the work known. Scanning at that speed requires slicing the spine off a book, and the pages are not put back. By October 2024 internal documents indicate millions of books had been bought. Two things in the viral telling go beyond the record: no confirmation exists that other AI companies did the same, and whether the targeted books were rare in any collector's sense is unclear. And one thing tends to get dropped: a federal judge ruled the practice fair use. This file separates what the documents show from what the posts added.
Read the case file →Viral online culture is now largely fake, secretly manufactured by AI influencers, synthetic personas, and coordinated engagement
The claim that online virality has quietly become fake: that AI-generated influencers, synthetic video and voice personas, bot armies, and paid engagement farms now secretly manufacture the trends, 'viral' moments, and popular accounts people see, so that little of what trends is organic. Much of the documented core is true. Openly AI-generated influencers earn real money and real followings, mass-produced 'AI slop' is measurably flooding feeds, and platforms have removed enormous numbers of fake accounts and engagement. What overreaches is the strong version, that essentially all of it is centrally coordinated and that organic human culture online is gone. This file separates the evidenced trend from the sweeping conspiracy. It is a close cousin of the older, broader Dead Internet Theory, narrowed to the specific 2025-2026 claim about manufactured virality.
Read the case file →Artificial intelligence, from chatbots like Grok to AI deepfakes, is being used as part of a coordinated covert scheme to secretly control or rig the 2026 US midterm elections
Going into the 2026 US midterm elections, a claim has spread that artificial intelligence is the hidden instrument of a plot to rig the vote: that AI chatbots such as Grok, together with AI-generated deepfake audio and video of candidates, are being coordinated to covertly control who wins. The claim braids together two things that should be kept separate. One is documented and real: generative AI tools have a well-recorded habit of producing false election information, and synthetic media impersonating candidates has genuinely circulated, both of which election officials and security agencies treat as serious threats to information integrity. The other is the rated conspiracy: that this amounts to a single coordinated covert scheme to secretly predetermine the 2026 result. This file states the record plainly and with sources, then explains why the leap to a centralized rigging operation is, on the public evidence, unproven. It attributes every allegation as an allegation, names no culprit, and accuses no one of a crime.
Read the case file →The video going round proves what happened: authentic footage circulating during a breaking news event documents the event it is captioned with
The advice of the last three years has been to watch for the tells of synthetic media: hands with too many fingers, dissolving text, impossible physics. That advice works, and it has trained a large audience to check the wrong thing. The most effective misinformation during a breaking news event is not generated at all. It is authentic footage, filmed somewhere else or some years ago, with a new caption attached. Nothing about the pixels is wrong, so every instinct a careful viewer has developed returns nothing, and the absence of a tell reads as a pass. During one week of the Ceuta border crisis in July 2026, fact-checkers traced clips captioned as the crossing to riots in Nador in Morocco, to Libya in 2023, to a commemoration of the prophet Noah on a Turkish mountain, and to the earlier 2021 Ceuta crisis. In the United States, footage of an apartment complex water fight circulated for months labelled as a security guard before being recaptioned as an ICE officer confronting children. This file rates the inference that footage passing an authenticity check documents the event it is captioned with, and on the record collected here that inference is debunked.
Read the case file →The footage going round after a disaster shows the disaster: clips circulating in the first hours after a flood, quake or landslide document the event they are captioned with
On 26 August 2026 a landslide and flash flood struck the Nepal-China border. Full Fact recorded more than 547 dead two days later; CBS News, cited by Snopes, put the toll at almost 600 with around 2,500 missing; a week later Reuters put the Nepali dead at 1,114, with nearly 4,000 missing across both sides of the border. Within hours the internet filled with footage. Almost none of it was from Nepal. Over the following six days Lead Stories, Fact Crescendo, Full Fact, Factly and Boatos published roughly twenty debunks between them: an AI-generated bridge collapse, before-and-after damage images made with OpenAI tools, a fighter jet supposedly causing the floods whose source clip had been online since May carrying an AI watermark, women staged as flood victims, rescue footage from the Congo in 2023, a landslide in Japan, an excavator swept away in Uttarakhand, a man trapped in debris in Prayagraj, a Buddha statue from 2024, a child rescue from 2024, and a flood scene lifted from an Indonesian television drama. Meanwhile Snopes rated something real. The CCTV of the water destroying the Gyirong Port crossing, which commenters under an X post with more than thirteen million views had confidently called AI, was authentic, and the Associated Press had reached the same conclusion. This file rates the assumption that footage in circulation during a disaster is footage of that disaster.
Read the case file →The quote is real and the photograph is real: statements and images circulating with a public figure's name on them are things they actually said and did
A screenshot of a quotation. A photograph of a politician in a T-shirt with a slogan on it. A clip of two familiar faces talking to children. Each arrives already attributed, needing nothing from the reader but recognition of the name. In six days in August 2026, fact-checkers rated eight of these across the United States and Britain and every one was manufactured. The techniques are what make the cluster worth a page rather than eight footnotes, because they have almost nothing in common: one photograph was altered, one was generated whole, one quotation was invented and put in a third party's mouth about the target, two were written as first-person statements the subjects never made, and one was an AI video of a conversation that never happened. Underneath sits a harder pattern than any single case shows. One account was found to have fabricated statements for the same person twice, months apart. Several producers openly label their output as satire and the label falls away at the first share. And the sexualised strand of this, which targets women in public life far more than men, is documented well enough to be counted: 100 incidents across 14 countries, 92 of them since 2022.
Read the case file →A sustained run of fabricated quotes, doctored videos, forged statements and fake photographs has been circulated about New York mayor Zohran Mamdani since his 2025 campaign, including a false claim that he intends to impose Sharia law
Zohran Mamdani was elected mayor of New York City on 4 November 2025, the first Muslim and first South Asian to hold the office. Around that campaign and the months since, an unusually sustained run of fabricated material has circulated about him, and the notable thing is how varied the techniques are. One was a straightforward impersonation: someone posing as former mayor Bill de Blasio gave The Times of London an interview criticising Mamdani, which the paper published, then retracted and apologised for within about ninety minutes of noticing. One was a forged terrorist communique, a fake Islamic State statement circulated to associate him with the group, debunked by AFP and terrorism analysts. Others are ordinary internet fakes: a doctored photograph placing a young Mamdani beside Jeffrey Epstein and Ghislaine Maxwell, a fabricated quote about reparations, a satire post recirculated as news, an AI video of him failing to applaud at a soldier's funeral. Running underneath them is a persistent claim that he means to impose Sharia law in New York, which PolitiFact examined and found baseless, and which is an anti-Muslim attack line rather than a description of anything in his record. This file rates whether these items are genuine. They are not. It takes no view on his politics, and it notes where a story attached to him turned out to be true.
Read the case file →Mitch McConnell is secretly dead, brain-dead, or incapacitated, and his condition is being covered up
In the summer of 2023, Mitch McConnell twice froze mid-sentence in front of cameras, unable to speak for many seconds. In June 2026 the 84-year-old senator entered a hospital after a fall, and for roughly a month his office declined to say why, offering only that he continued to improve. Into that silence rushed a familiar kind of theory: that McConnell was actually dead, brain-dead, or so incapacitated that aides were propping up a shell and hiding it from the public, some versions claiming a body double or an AI-generated photo. This case file keeps two things apart: on one side the real health scares, a real hospitalization, and a genuinely thin flow of information; on the other the claim we actually rate, that his death or replacement is being concealed. On the conspiracy, the evidence does not hold up: McConnell reappeared with a statement, a photo, and a third-party account of a phone call, and explained the hospitalization as a fall and mild pneumonia. The verdict is debunked, with the legitimate transparency question left standing on its own.
Read the case file →It really happened: stories that a satirical site published as jokes are circulating as reports of real events
In August 2026 two stories spread as news that had begun life as jokes. One said New York City's planned government-run grocery stores would serve anyone carrying a special mark on the hand or forehead, illustrated with the number 666. The other said Massachusetts Democrats had unveiled a statue of Molech, the deity the Bible associates with child sacrifice, outside the state Capitol. Snopes checked both and traced each to The Babylon Bee, a site that advertises itself as Christian news satire. Neither event happened. The interesting part is not that a satirical site published satire, it is what the two have in common: each takes a real and disputed policy argument and restates it as biblical prophecy, which is the form most likely to survive being stripped of its source. Research by Ohio State University found this is neither rare nor one-sided, with Babylon Bee stories among the most-shared inaccurate content in almost every wave of a six-month survey, and Onion stories fooling readers on the other side of the aisle. This file rates the circulating claims as false and does not take a view on the policies being satirised.
Read the case file →The screenshot proves they posted it: a circulating image of a public figure's social media post is reliable evidence that the post was made
A screenshot of a post looks like documentation. It is not; it is a picture, and pictures of text are trivial to make. The week of 25 to 31 August 2026 showed what that now costs. Lead Stories ruled that the president's account had not posted an image of him walking on water captioned Pastor, President, Protector; had not posted a mess hall picture, itself AI-generated, captioned about the military eating the best; had not announced a Melania Canal to bypass the Welland Canal, a fabrication traced to a satire account; and had not threatened to close New York's 9/11 memorial. In the same stretch it confirmed that a post about Kharg Island was genuine, and that it carried an AI-generated video of a strike no reporting supports. Snopes, meanwhile, published a collection titled around two dozen things the president actually posted, drawn from its archives, noting it routinely traces alleged posts to people with editing skills. Among the genuine items is an AI image of himself as the Pope. That is why the obvious test fails: when the real account posts outlandish and synthetic material, outlandishness stops being evidence of forgery.
Read the case file →Nigerian security forces are secretly supplying and evacuating the bandits they are sent to fight, and the helicopter videos prove it
A video appears. A helicopter sits in scrubland with armed men standing around it, and the caption says the security services are resupplying the bandits they are supposed to be fighting. The clip travels fast, because in a country where thousands have been killed and schoolchildren taken by the busload, it answers the question that will not go away: how can this still be happening. Every such video that reputable Nigerian fact-checkers and newsrooms have run down has resolved in one of two ways, and neither is a supply drop. Some were filmed in other countries and relabelled. The rest are real Nigerian security operations, with the aircraft doing surveillance for troops and vigilantes on the ground. What makes this more than a run of bad captions is where the suspicion comes from. In March 2021 the federal government itself declared Zamfara a no-fly zone and banned all mining there, and the reason briefed to reporters was that aircraft and illegal gold were feeding an arms trade. The state told the country that planes were part of the problem. The videos supply the pictures.
Read the case file →The crew of the USS Abraham Lincoln is being silenced and killed aboard a carrier the Navy will not bring home, and the president has ordered sailors to sign non-disclosure agreements and called them suckers and losers
In November 2025 the aircraft carrier USS Abraham Lincoln left San Diego on what was described as a routine deployment. It was redirected to the Arabian Sea after the war with Iran began, and it did not come home in May 2026 as planned. By July it had set a record for consecutive days at sea; by early August it had been out more than 259 days with a crew of around 5,000. Reporting from the BBC, Stars and Stripes and CNN has described food shortages, broken plumbing and mounting strain, and lawmakers have asked questions; the Defense Secretary dismissed an earlier round of the same reporting as fake news, and the Navy says it has not observed an increase in suicidal ideations or attempts aboard. That argument is real, and this file does not settle it. What this file rates is the cluster of fabrications that grew on top of it in a single week of August 2026: a satirical post about mandatory non-disclosure agreements that lost its label and was taken as news, a forged screenshot of a presidential post calling servicemembers suckers and losers, and a claim from Iranian state media that seven sailors had died in a brawl, which US Central Command denied. Those three are the claim rated here, and they do not hold up. Snopes has since published a standing collection of the Lincoln claims it has examined, which is a fair measure of how much has attached itself to one ship.
Read the case file →Videos circulating online prove Moroccan border guards deliberately opened the frontier at Ceuta in July 2026 to engineer a mass crossing into Spain
Starting on Thursday 30 July 2026, tens of thousands of people crossed from Morocco into Ceuta, the Spanish city on the north African coast, mostly by swimming around the breakwaters at Tarajal and Benzu. Spain's Interior Ministry put the figure at close to 50,000 in roughly a day; Ceuta's president said 60,000. Dozens died. It was, by any measure, a major event, and within hours videos were circulating captioned as proof that it had been deliberately opened from the Moroccan side. Two of the most shared were checked by the fact-checking outlet Lead Stories on 31 July and neither is what it claims. One, captioned as showing a Moroccan guard opening the border fence in July 2026, was posted online in 2021. The other, shared with the caption 'Is this an invasion of Spain?', had been online since at least 5 July under a caption that mentioned neither country, and two AI detection tools rated it likely machine-generated, with visible anatomical errors in the crowd. This file keeps the layers apart. The crossings, the deaths, and the emergency are documented and are not in question. The claim rated here is that these particular clips prove a staged operation, and that claim is debunked. The larger question of what Moroccan enforcement did or did not do in July 2026 is unresolved, and is treated as such below.
Read the case file →The camera does not lie: a viral clip of an argument, a theft or a confession is footage of something that happened
A creator films a sketch, writes Comedy on it, and posts it. Someone downloads that file, throws away the caption, and reposts it as a real fight, a real theft, a real confession. The video is unchanged; the lie is everything around it. Six checks over twelve months show how routine this has become. Factly traced a mid-flight altercation to Fly High Institute Nagpur, an air-hostess training school that had posted it in three parts as cabin-crew training material. It traced a woman's confession that she had married her father-in-law to a sketch channel whose videos carry entertainment disclaimers. It traced a chain-snatching to a creator whose caption read, in full, that the video is totally scripted and made for entertainment and awareness. It traced two communal videos, one of a man mocking an elderly Muslim man and one of a woman destroying a saffron-coloured motorcycle, to a comedy duo and to a prank account. Lead Stories traced a shelter video about an 83-year-old surrendering her dog to an account whose original upload said it was made with AI for entertainment and storytelling. In every case the label existed. In every case it was outside the video file, and so it did not travel.
Read the case file →They really said it, so the quote is fair game: a statement confirmed as genuine and correctly attributed can be passed on as evidence of what its speaker thinks
The instinct to ask whether a quote is real is a good one, and it is now widespread enough that fact-checkers answer it constantly. What almost nobody asks is the second question. In 2026 Snopes rated a run of quote claims Correct Attribution, its verdict for statements that were genuinely made by the person credited, and in case after case the confirmation did not settle the matter. A Florida meteorologist's line about no longer being able to predict hurricanes was exactly as quoted, from 3 June 2025, and was circulating in August 2026 as though it described the season then underway. Sean Penn's description of a president as an enemy of mankind was his, written in a 2018 Time essay about relief work in Haiti after the 2010 earthquake, and spread eight years later as a fresh denunciation. A Heritage Foundation chairman's 2021 remarks were correctly attributed for most of their length, with someone else's commentary on them attached to the end. Martina Navratilova's reply was real, made on social media to a video rather than to the president in a room. This file rates the assumption that a verified quote is a settled one.
Read the case file →Much of the viral “war footage” shared during the 2026 Iran conflict was not real: fact-checkers traced widely circulated “strike” and “missile” clips to the video game Arma 3, a 2015 warehouse explosion in China, an Algerian football celebration, and AI-generated imagery
As Iran, Israel and eventually the United States were drawn into open confrontation across 2025 and 2026, social feeds filled with dramatic clips captioned as live footage of strikes, missile barrages and downed jets. A great many were not what they claimed to be. Fact-checking teams at CNN, Poynter, NewsGuard, BBC Verify, Euronews, PolitiFact and others independently traced the most viral examples to unrelated origins: the war game Arma 3, a 2015 warehouse disaster in China, a 2021 Ukrainian depot fire, an Algerian football celebration, and a swelling stream of AI-generated imagery. Some clips had already been debunked in the Ukraine, Gaza and India-Pakistan contexts before being recycled here. This file is a case study in war disinformation: it reports the documented fakes and how they were spotted, rates those specific clips debunked, and stays strictly neutral on the conflict itself, taking no side and attributing every claim to the body that established it.
Read the case file →The announcement is official: a graphic carrying a ministry's, a broadcaster's or a company's house style speaks for that institution
Three years of public advice about misinformation have pointed at faces. Watch the hands, watch the mouth, watch for the shimmer where a jawline meets a collar. The artifacts collected here contain no faces to inspect. They are announcement cards: a ministry advisory, a school reopening notice, an arrest, a candidate's withdrawal, a vaccine release, a price. Each borrows the visual grammar an institution uses for its own genuine announcements, and each was fabricated. In the week of 17 August 2026 alone, Africa Check rated four of them across Kenya and South Africa, Lead Stories rated four counterfeit news outlets across the United Kingdom, Germany, Italy and the United States, and Boatos rated a Brazilian public health message built around the mascot of the national immunisation programme. What connects them is not a technology. It is that institutions now publish real announcements in exactly this form, as a shareable image with a logo on it, so the counterfeit and the genuine article are the same object. This file rates the assumption that the house style is the proof, and on this record it does not hold.
Read the case file →Authentic photographs, video, and audio of public figures are AI fabrications, so any inconvenient recording can be dismissed as fake
As deepfakes got good, a second problem arrived behind the first. Once the public knows convincing fakes exist, anyone caught on camera can simply say the camera lied. Legal scholars Bobby Chesney and Danielle Citron named this the liar's dividend in 2019, and made the uncomfortable observation that it grows in proportion to how well people are educated about deepfakes: every explainer about how convincing fakes have become is also an argument that nothing can be trusted. This case file rates the specific claim, made repeatedly and in public, that authentic images and recordings are AI fabrications. It holds that claim up against what happened when each one was checked.
Read the case file →The recording proves what they really said: leaked audio of a public figure is authentic, because you can hear their own voice saying it
For most of the last century a recording of somebody's voice was close to conclusive. Tape could be spliced and quotes could be trimmed, but the voice itself was the part nobody could fake. That is no longer true, and the change arrived faster than the habit of trusting it could adjust. Two days before Slovakia's 2023 parliamentary election, audio circulated that seemed to capture a party leader and a journalist planning to rig the vote; both said immediately it was fabricated, AFP found numerous signs of manipulation, and it went out during a 48-hour moratorium when Slovak media and politicians are required to stay quiet. Four months later thousands of New Hampshire households answered the phone to an AI-generated Joe Biden telling them not to bother voting in the primary. By August 2026 the technique had been simplified: instead of generating a whole recording, fabricated speech is now dropped onto authentic video, so the footage passes every check while saying something its subject never said. This file rates the assumption underneath all of it, that hearing the voice settles the question.
Read the case file →Viral AI-generated images showing Jeffrey Epstein alongside politicians and celebrities are real photographs leaked from the government's Epstein files
When the Justice Department released millions of pages of Epstein-related records in early 2026, the genuine disclosure was quickly polluted by a flood of fakes. AI image generators were used to manufacture realistic-looking 'photographs' of Jeffrey Epstein posed with prominent political and celebrity figures, which then circulated as if they were leaked evidence from the files. Open-source researchers and fact-checkers, Open Measures, Snopes, Bellingcat, and the European Broadcasting Union's forensic team among them, examined the most viral examples and identified them as synthetic, not as anything found in the government release. This case file rates the claim that these viral AI images are authentic leaked photos. It is false. It also covers the second-order damage: a real archive of documents becomes harder to discuss when the information space is saturated with convincing fabrications, and the mere existence of deepfakes hands the genuinely guilty a way to wave off authentic evidence.
Read the case file →Caught on camera: the confrontation videos circulating as real security footage and bystander clips are documentary evidence of things that happened
A clip arrives with no context and a caption in the first person. A landlord forced her way into my home. A father knocks out the man who followed his daughter from school. A police officer refuses to pay for a coffee. The framing is always the same: this is footage, somebody's camera happened to be running, and you are seeing what occurred. In four cases checked during a single week in August 2026, across the United States, Thailand and Brazil, it was not footage of anything. It was scripted entertainment, made by production companies and skit channels that labelled their own work, and then stripped of that labelling by whoever posted it next. One case shows the step happening: the original carries a studio watermark in the corner, and the viral repost has the watermark blurred out. Running alongside it is a stranger second genre, in which AI generates fake behind-the-scenes footage of film crews building disasters on miniature sets. That one matters more than it looks, because it manufactures precisely the artifact that every staged-event conspiracy theory has always asked for and never had.
Read the case file →India's Cockroach Janta Party protests were not an organic student movement but a foreign-backed influence operation, run with Pakistani bot followers, US funding, and an opposition toolkit
In May 2026 the Chief Justice of India, Surya Kant, said during a Supreme Court hearing that there were 'youngsters like cockroaches' who could not find employment, and referred to 'parasites' attacking the system. Within a day a political communications strategist, Abhijeet Dipke, had registered a satirical website calling itself the Cockroach Janta Party, styled as the voice of the lazy and unemployed. It was a joke that stopped being one. Fed by allegations that the NEET-UG medical entrance paper had leaked and by the cancellation and re-sit that followed, the CJP became a mass movement, out-followed the governing party on Instagram, filled Jantar Mantar from 6 June, and ended on 25 July with the education minister's resignation. Running alongside it the whole way was a second story: that none of this was real. Versions held that the followers were Pakistani bots, that the money was American, that an opposition leader was secretly behind the account, and that a leaked toolkit showed the outrage was manufactured. This file keeps the two apart. That the protests happened, and that this counter-narrative was pushed at them, are both documented. The counter-narrative itself is what gets rated, and each of its load-bearing claims has been checked and found false.
Read the case file →Viral videos show Muslims taking over British streets and transport, blocking commuters, shutting down London, and forcing halal rules on the country
Since spring 2026 a steady run of videos has circulated in Britain appearing to show Muslims occupying public space: praying in the middle of London streets, marching in their thousands to demand halal food standards, and kneeling on prayer mats down the aisle of a London Underground carriage while trapped commuters complain that they cannot get off. The footage looks convincing, the captions are furious, and the comment sections fill with people who believe they are watching the news. None of it happened. Full Fact has debunked at least four of these items between April and August 2026, and in each case the video was generated by artificial intelligence or the report was assembled from unrelated old photographs. The Tube clip was rated 99.9 percent likely AI by one detection tool; the halal march carried a Google SynthID watermark in both audio and visuals. This file rates the genre, not a single clip, and the rating is that it is fabricated. It is worth being plain about what these videos are for: they invent misbehaviour by a real minority in order to make people angry at their neighbours. The most useful thing about them is how they fail, which is documented below.
Read the case file →A video circulating since 5 August 2026 shows the moment a SpaceX rocket stage slammed into the Moon, throwing up a plume and leaving a hundred-foot crater
On 5 August 2026 a spent SpaceX upper stage hit the Moon. Astronomers had seen it coming, and the European Southern Observatory confirmed the impact the way such things are usually confirmed: spectroscopically, by detecting sodium and lithium in a plume that lasted a few minutes. Nobody produced footage, because confirming an impact and filming one are different problems. Within fifteen minutes of the predicted moment, a video appeared on TikTok showing an object striking the lunar surface and throwing up a white cloud, and it went on to millions of views on X, with one caption claiming a hundred-foot crater. The clip is not from 5 August. Snopes reverse-searched it to a wider version posted in April, four months earlier, by a Facebook page that produces AI-generated videos of mysterious objects hitting the Moon. Full Fact asked astronomers, who said it does not match the Moon's real terrain or the true impact site. This file keeps the two things apart: a real, predicted collision confirmed without pictures, and a synthetic video that arrived to fill the gap where the pictures would have been.
Read the case file →A July 2026 wave of AI deepfake videos falsely showed Union Minister Piyush Goyal and Prime Minister Narendra Modi threatening student protesters as though those were their real words; the most viral clip took genuine PTI footage of Goyal and swapped in a fabricated audio track, and the government's PIB Fact Check unit dismissed the videos as AI-generated, with an FIR registered on 25 July
In late July 2026, as students protested at Jantar Mantar over an alleged NEET-UG paper leak, a batch of short videos spread that appeared to show senior Indian leaders threatening the demonstrators. In the most viral, Union Minister Piyush Goyal seemed to call, on camera at a real media interaction, for protesters to be violently punished; another clip put similar words in Prime Minister Narendra Modi's mouth. None of it was authentic. The Goyal video was built from genuine footage recorded by PTI outside Parliament on 23 July, with the original audio stripped out and an AI-generated track dropped in, so the visuals looked like a real news clip while the words were invented. The government's PIB Fact Check unit dismissed the Modi and Goyal videos as AI-generated deepfakes. Goyal said his remarks had been maliciously doctored, an FIR was registered on 25 July, and platform links were taken down. This case file treats the videos as what fact-checkers found them to be, manufactured clips, and examines how an audio swap on real footage can weaponise a public figure's own image within hours. It does not repeat the fabricated remarks as anyone's actual words.
Read the case file →The European Union is building a censorship grid over the internet: an 'internet passport' will be required before anyone can go online, VPNs are to be banned so nobody can escape it, and the AI Act completes the control system
Through 2026 a set of claims spread across European and English-language social media holding that Brussels was about to close the open internet: that an 'internet passport' would soon be required before anyone could go online, that VPNs would be banned so the requirement could not be evaded, and that the AI Act supplied the censorship machinery to finish the job. Individual posts were viewed millions of times. Each claim points at something real and then misdescribes it. The passport is an age-verification app the Commission is building so that platforms can keep children away from age-restricted material; a user verifies their age once with an official document and afterwards proves only that they are above a threshold, and it does not gate the internet. The VPN ban does not exist: a Commission official said the goal is making age checks harder to circumvent, and her office said flatly there is no crackdown on VPNs and no proposal targeting them. The AI Act obligations that took effect on 2 August 2026 are a disclosure rule requiring deepfakes to be labelled as artificial, not a takedown power. There are real arguments to have about all three, and digital-rights groups are having them. This file rates the viral version, and the viral version is false.
Read the case file →A viral video showing President Trump collapsing as he got into a vehicle is real and proves his declining health is being hidden
In late July 2026 a short video spread across X and TikTok that appeared to show President Donald Trump, then 80, collapsing as he tried to get into a vehicle. Many who shared it presented it as proof that his declining health was being concealed from the public. The footage is not real: it was generated with AI software, and the account that first posted it wrote in a caption that the video was AI even as others reshared it as authentic. The fake was built from genuine footage published on 1 July 2026 that shows Trump pausing briefly, one foot in the doorway, before completing an unremarkable step into the car; at no point in the real clip does he fall. Fact-checkers including Snopes and Lead Stories examined the viral version, flagged tell-tale signs of AI synthesis such as blurred edges and misshapen hands, and found no credible news report of any such collapse. This case file keeps two things apart: the president's age and the ordinary real footage the fake was derived from; and the claim it rates, that the collapse video is genuine. That claim is debunked. It is a textbook synthetic-media hoax: a real, mundane moment re-rendered by AI into a dramatic one, then circulated as evidence.
Read the case file →Videos showing famous people and government officials personally endorsing an investment platform, crypto giveaway, or miracle product are genuine recordings of those endorsements
Most synthetic-media stories are about politics. This one is about money. Since roughly 2023 the same tools that generate political deepfakes have been aimed at a simpler goal: producing a video of someone recognizable, a film star, a business figure, a finance minister, appearing to endorse an investment platform, then buying advertising to put it in front of people. The FBI's Internet Crime Complaint Center catalogued the technique in a December 2024 public service announcement, listing AI-generated celebrity images used to promote products and AI-generated video used for promotional material in investment fraud, alongside cloned voices and fabricated websites. The pattern recurs across jurisdictions: in July 2026 India's Press Information Bureau debunked an advertisement showing Finance Minister Nirmala Sitharaman apparently endorsing a high-return scheme, one of several versions in which the promised returns were swapped between uploads while the pitch stayed the same. This file rates the claim that such endorsements are genuine. They are not, and what gives them away is rarely in the pixels; it is in the offer.
Read the case file →Videos circulating since late 2025 that appear to show Ukrainian soldiers weeping, apologizing to the Russian people, and surrendering en masse are authentic front-line footage
In early November 2025, videos began circulating that appeared to show Ukrainian soldiers at their lowest: weeping, calling for their mothers, apologizing to the Russian people, blaming their own government for the war, and in some clips surrendering to Russian forces in groups. The footage was ultrarealistic and it traveled fast across YouTube, TikTok, Facebook and X. It was also generated. NBC News reviewed 21 of the videos and reported on 13 December 2025 that in at least half the giveaway was not subtle at all: Sora 2 stamps a small logo on downloaded video, and it was still there. Clips that appeared to come from other generators carried quieter errors instead, principally uniforms and helmets that were incorrect or simplified versions of real Ukrainian kit. A third signal was plain to Ukrainian viewers and meaningless to everyone else: most of the soldiers spoke Russian, and only eight of the 21 clips featured anyone speaking Ukrainian. One recurring face belonged to Aleksei Gubanov, a livestreamer who had fled Russia over his opposition to Putin and recognized himself in a uniform he had never worn. This file separates the real war, in which soldiers genuinely are captured and killed, from the rated claim that these particular videos document it. That claim is debunked. Who made and seeded them has not been established.
Read the case file →The viral “CCTV” clips said to show the thieves inside the Louvre during the October 2025 crown-jewels heist are not real security footage: fact-checkers traced them to OpenAI’s Sora 2 video generator
Hours after masked thieves stole crown jewels from the Louvre’s Apollo Gallery on 19 October 2025, feeds filled with grainy, timestamped clips presented as CCTV or phone video of the raid in progress. They looked like the leaked footage everyone wanted, and they were fakes. Fact-checking teams at France 24, AFP, the South China Morning Post and CEDMO independently traced the clips to OpenAI’s Sora 2 generator, spotting a Sora watermark left visible in an early upload, a second watermark pasted over it to hide the tell, a filming hand reflected in empty space, and a display-case layout that does not match the real gallery. No authentic interior footage of the heist was ever released, and the vacuum was filled with synthetic video. This file keeps the two layers apart: the robbery is a documented crime, and the viral “proof” of it is AI-generated. It rates only the clips, and rates them debunked.
Read the case file →Viral photos of a Zendaya and Tom Holland wedding at Lake Como are real, leaked images that document the couple's secret marriage
In March 2026, a run of strikingly realistic images showing Zendaya in a satin wedding gown and Tom Holland in a tuxedo, most of them set beside Italy's Lake Como, raced across social media and gathered more than ten million likes. Many people took them for leaked photographs of a secret wedding. They were AI-generated. Zendaya confirmed as much herself on Jimmy Kimmel Live, saying that many people had been fooled, that acquaintances had congratulated her on the beautiful pictures, and that she had to tell them the images were AI. Her stylist Law Roach, the person who actually dresses her, brushed the fabricated gown aside as not good enough. This case file separates two things the story kept fusing: what is on the record, that Zendaya and Tom Holland are a genuine, publicly known couple, reported to be engaged, and the claim it rates, that the viral pictures are real photographs of their wedding. The relationship is real. The photos are not. On that claim the verdict is debunked, and the episode stands as one of the cleanest media-literacy cases of the year: convincing synthetic images of a plausible event, believed by millions before anyone paused to check.
Read the case file →Many viral 'moments' from the 2026 FIFA World Cup were fabricated: an AI image of Benjamin Netanyahu in an Argentina shirt, a face-swapped 'Haaland reacting to his reflection' clip, invented player backstories and fake quotes were all passed off as real
The 2026 World Cup was the first men's tournament held after cheap, photorealistic AI video and image tools reached the mainstream, and it showed. Alongside real matches, feeds filled with clips and pictures presented as genuine World Cup moments that were nothing of the kind: an AI image of Benjamin Netanyahu cheering Argentina from the stands (he was hosting meetings in Jerusalem that day), a viral 'Haaland reacting to his reflection' video that turned out to be a face-swapped Chinese comedy skit, invented player quotes and backstories, and, after Spain beat Argentina in the final, fabricated images used to claim the result was rigged. Fact-checking teams at Lead Stories, Euronews' The Cube, the European Broadcasting Union, AFP, Misbar, NewsGuard and public broadcasters traced each to its real origin, often using reverse image search, provenance checks and watermark detectors such as Google's SynthID and OpenAI's verification tool. This file catalogues the documented fakes and keeps the real tournament and its results strictly separate from the synthetic media that rode on its back.
Read the case file →Viral videos showing dramatic ICE raids at Walmart, McDonald's, and other workplaces are authentic footage of real immigration enforcement
Through late 2025 and into 2026, hundreds of AI-generated videos depicting supposed ICE immigration raids, at Walmarts, McDonald's, workplaces, schools, and on city streets, racked up millions of views across Facebook, Instagram, and YouTube. Engagement accounts such as 'USA Journey 897' mass-produced them with text-to-video tools, principally OpenAI's Sora. Fact-checkers including Full Fact, HuffPost, Euronews, and Snopes confirmed the specific viral clips are synthetic fabrications, not footage of real events, identifying them through leftover Sora watermarks, uniform clip lengths, and visual errors. This case file debunks the authenticity of those clips only. It takes no position on actual immigration enforcement, which is documented separately by conventional journalism. The deeper story is the 'liar's dividend': a flood of convincing fakes does double damage, spreading false events while also giving anyone cover to dismiss genuine footage as 'just AI.'
Read the case file →