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
Where the evidence lands: FalseThat the widely shared clips captioned as CCTV or bystander footage from inside the Louvre, appearing to show the thieves cutting into display cases and grabbing the crown jewels on 19 October 2025, are authentic recordings of the real robbery.
Believed by: Millions of social-media users who watched and reshared the clips before they were labeled; one version alone was posted by a Hong Kong page with more than 240,000 followers, and the clips were widely mistaken for leaked or official security footage
Reviewed and rated debunked. The viral clips presented as CCTV or bystander footage of the Louvre crown-jewels heist are AI-generated with OpenAI’s Sora 2, per France 24, AFP, the South China Morning Post and CEDMO: an early upload still carried the Sora watermark, a second watermark was pasted over it to hide the tell, and the clips show a filming hand reflected in empty space and a display-case layout that does not match the real Apollo Gallery. The robbery itself, about 88 million euros of jewels stolen on 19 October 2025, is real and documented; only the “footage” is fake. source →
The full story
The real heist, and the footage that never existed
Keep two things apart from the start. On 19 October 2025, about half an hour after the Louvre opened, a crew disguised as workers drove a truck-mounted lift against the museum’s Seine-side facade, cut through a window into the Apollo Gallery, forced two display cases and made off with eight pieces of the French crown jewels, worth roughly 88 million euros. The raid took only minutes. That robbery is real, documented by wire services and French authorities, and it is not what this file rates.
What it rates is the wave of clips that appeared within hours, captioned as CCTV or phone footage from inside the theft. They had the look everyone wanted: grainy, timestamped, shot from a fixed overhead angle, the visual grammar of a leaked security tape. They spread across TikTok, Instagram and Facebook as the thing the story seemed to be missing. And they were fakes.
The question here is narrow and answerable: were these clips real recordings of the robbery? No. Independent fact-checkers traced them to a video generator. The heist happened; the “video of it” did not.
Why the fakes were so easy to believe
It is worth stating plainly why so many sensible people shared these clips, because the reasons are not stupidity. First, the event was genuinely staggering: a daylight raid on the most famous museum on earth, crown jewels gone in minutes. A viewer’s reasonable prior was that dramatic footage of such a thing must exist, so the appearance of a clip felt like confirmation, not a claim to be tested.
Second, the tool was good. These clips were made with Sora 2, OpenAI’s video generator, and they carried exactly the texture people read as authentic: low resolution, a burned-in timestamp, the still, unglamorous angle of a ceiling camera. Nothing looked produced, and that plainness is persuasive.
Third, and most important, there was a real vacuum. No authentic interior footage of the robbery was released, because security recordings from an open criminal case do not go on social media. Curiosity had nowhere legitimate to land, so the first plausible “heist video” to surface absorbed all of it. The fake did not have to beat the real footage; there was no real footage to beat.
The tells: watermarks, reflections and a gallery that is wrong
The debunk is clean because several teams reached the same place. The decisive piece was provenance. Using a reverse image search, AFP surfaced an early French-language TikTok upload in which the “Sora” watermark, OpenAI’s generator tag, was still visible on the footage. In another version, spread by a Hong Kong page with more than 240,000 followers, a “Websiren” watermark appeared to have been laid over the clip precisely to cover that Sora tag, a paste-over that is itself a sign of manipulation, not authenticity.
Then the physics. AFP flagged that the hand of the person supposedly filming was reflected in empty space, with no mirror to justify it, and that a burglar appeared to step onto a spot with nothing to support his weight. Reviewers also noted that the display-case layoutin the clips did not match the real Apollo Gallery, and that separate versions of the “footage” depicted the raid in mutually contradictory ways, different thieves, different methods, different numbers, which is what you get from a generator, not from a camera.
A generator’s watermark, a second watermark pasted over it, a hand reflected in nothing, and a gallery that is not the real one. No camera recorded any of it.
Why fake footage of a real crime spreads
The pattern is more interesting than gullibility. When a real, high-interest event has no available footage, it creates a demand that generative tools are now perfectly positioned to meet. The audience is primed to believe something dramatic happened, because it did, and the missing clip is the one thing the story seems to lack. A synthetic video does not have to convince from scratch; it only has to fit an expectation the news already built.
Layer on the ordinary economics of feeds, where the account that posts the sensational thing first is rewarded with reach, and the incentive to manufacture and repost becomes structural. Add tools that can produce a convincing security-camera clip in minutes, and the supply of plausible fakes is effectively unlimited. None of this requires a grand plot. It needs a real event, a footage gap, cheap tools and a fast feed.
The harm runs two ways. The fakes mislead directly, and they also corrode trust in the genuine record: once people know clips can be invented, real evidence can be waved off as “probably AI.” Guarding against both credulity and reflexive denial is the whole job.
Where the evidence lands
Hold the two layers apart, because that is the entire point. The heist was real: about 88 million euros of crown jewels were taken from the Apollo Gallery on 19 October 2025, an event reported by wire services, confirmed by French authorities, and followed by arrests and charges. This file says nothing to cast doubt on that.
The viral clipspresented as footage of that theft are a different matter. France 24, AFP, the South China Morning Post and CEDMO independently traced them to OpenAI’s Sora 2 generator, with a visible Sora watermark, a pasted-over cover watermark, impossible reflections and a wrong gallery layout sealing it. No authentic interior footage of the robbery ever circulated. On those clips the rating is debunked.
The lesson is portable. The next big, camera-shy event will bring the same reflex: a real story, a footage gap, and a generator ready to fill it. The defense does not change. Check provenance before virality, look for the tells, and remember that the absence of real footage is an invitation to fakery, not a reason to trust the first clip that answers the wish.
What's still unexplained
- How can a viewer check a clip like this? The methods that worked here were ordinary: reverse image search to find the original upload (which exposed the Sora watermark), looking for provenance tells such as pasted-over watermarks, and asking whether reputable newsrooms carry the same interior footage. If only social accounts have it, treat it as unverified.
- Why did the vacuum form? Genuine security footage from an active criminal case is rarely published, so a real, high-interest event can go without any authentic video for a long time. That gap between public curiosity and available footage is the opening synthetic media exploits.
- Do watermark tricks defeat detection? Partly. Overlaying a second watermark to hide a generator’s tag, as happened here, can slow casual identification, which is why provenance and physical-consistency checks, not a single visible logo, remain the reliable test.
Point by point
The claim: The grainy, timestamped clips are real CCTV or phone footage of the thieves inside the Apollo Gallery.
What the record shows: They are AI-generated. France 24’s Truth or Fake, AFP, the South China Morning Post and CEDMO independently attribute the clips to OpenAI’s Sora 2. A reverse image search by AFP surfaced an early French-language TikTok upload in which the “Sora” watermark, OpenAI’s generator tag, is still visible on the footage.
The claim: A watermark on the video shows it came from a real camera system.
What the record shows: The opposite. Fact-checkers found a “Websiren” watermark appeared to have been pasted over the clip specifically to cover the Sora AI watermark, which still showed through in at least one of the posts where the video first appeared. The overlay is a tell of manipulation, not a mark of authenticity.
The claim: The footage looks physically real, so it must be genuine.
What the record shows: It fails on physics and geography. AFP noted the hand of the person supposedly filming is reflected in empty space rather than in any mirror, and a burglar appears to step onto a spot with nothing to support his weight. Reviewers also noted the display-case arrangement in the clips does not match the real Apollo Gallery, and separate versions depict the raid in mutually contradictory ways.
The claim: So many pages and people shared it, including big accounts, that it has to be real.
What the record shows: Reach is not verification. One version was spread by a Hong Kong page with more than 240,000 followers, which is exactly how a plausible fake travels. Prominent resharers and high view counts are what synthetic media exploits, not evidence that a camera recorded anything.
The claim: There must be real inside footage, and this is it.
What the record shows: No authentic interior footage of the robbery circulated publicly. The Louvre’s own security recordings were part of a criminal investigation, not a social feed. That absence is precisely the gap the Sora clips rushed to fill, offering a synthetic “video” where no real one was available.
The claim: If the footage is fake, the whole heist was staged or never happened.
What the record shows: That inference is its own error. The theft is a documented crime: about 88 million euros of crown jewels were taken from the Apollo Gallery on 19 October 2025, reported by wire services and confirmed by French authorities, with arrests and charges following. Debunking the clips establishes only that the clips are fake, nothing about whether the robbery occurred. It plainly did.
Other readings
Angles that don't fit neatly into the claim or its rebuttal, laid out and weighed, not endorsed.
Part of a wider pattern of AI “footage” of real events
The Louvre clips are not a one-off. The Conspiratory separately documents how much of the viral “war footage” shared during the 2026 Iran conflict was AI-generated or miscaptioned, traced by fact-checkers to game engines, recycled disasters and synthetic video. The Louvre case is the same mechanism on a crime story: a real, dramatic event, no available authentic footage, and a generator ready to manufacture the clip the audience wants. See the file on the 2026 Iran war fakes for the broader playbook.
The “staged heist” tangent, and why it is separate
A different online current treated the robbery itself as suspicious, framing it as a stunt or inside job. That is a distinct claim from the one rated here. This file addresses only whether the viral clips are real footage (they are not); it takes no position on the wider speculation, which the documented investigation, arrests and charges do not support.
Timeline
- 2025-10-19About half an hour after the Louvre opens, a crew disguised as workers parks a truck-mounted basket lift against the Seine-side facade, cuts through a window into the Apollo Gallery, smashes two display cases with power tools and makes off with eight pieces of the French crown jewels, valued at roughly 88 million euros. The raid lasts only a few minutes; the thieves flee on scooters. This is the real, documented event.
- 2025-10-19Within hours, clips styled as security-camera or phone footage of the theft begin circulating on TikTok, Instagram and Facebook. No genuine interior footage of the robbery has been released, but the demand for “the video” is immediate and enormous.
- 2025-10-20Different versions of the “heist footage” spread that contradict one another, showing the thieves, their number and their methods in incompatible ways: a sign that the clips are generated, not recorded. Several carry a grainy, timestamped look that reads as authentic CCTV.
- 2025-10-23France 24’s Truth or Fake publishes a segment walking through the circulating clips and concludes they are AI-generated with OpenAI’s Sora 2, not real footage of the robbery.
- 2025-10-24AFP’s forensic debunk, echoed by the EU disinformation hub CEDMO, reports that a reverse image search surfaced a French-language TikTok version of the clip in which the “Sora” logo is still visible, and flags visual anomalies: a filming hand reflected in empty space and a burglar stepping to a spot with nothing to support his weight.
- 2025-10A Facebook reel of the fake footage is posted by a Hong Kong-based page with more than 240,000 followers; the South China Morning Post fact-checks it, noting a “Websiren” watermark appears to have been laid over the video to cover the original Sora AI watermark.
- 2025-11-25The real investigation continues: French authorities announce further arrests, with several suspects charged over the heist. The stolen jewels remain unrecovered, underscoring that the crime is genuine even as its “footage” is not.
- 2026-07-22The Apollo Gallery reopens to the public, its cases empty of the stolen jewels. By this point the Sora clips have become a standard classroom example of how convincing AI video can fill an information vacuum around a real event.
False. This file rates the viral “footage,” not the robbery. The Louvre heist itself was real: on 19 October 2025 thieves stole about 88 million euros of French crown jewels from the Apollo Gallery, a documented event that is not in question here. What is rated is the wave of clips captioned as authentic CCTV or bystander video from inside the theft. Working independently, France 24’s Truth or Fake, AFP, the South China Morning Post and the EU disinformation hub CEDMO traced the most-shared examples to OpenAI’s Sora 2 video generator, not to any camera in the museum. The clips carry AI tells (a Sora watermark left visible in an early upload, a filming hand reflected in empty space, a thief stepping onto nothing, a display-case layout that does not match the real gallery) and no genuine interior footage of the robbery ever circulated. On those clips the verdict is debunked. This is not a claim that the heist did not happen; it is the narrower, sourced finding that the “video of it” was machine-made.
Reviewed by The Conspiratory Editors · Last reviewed July 27, 2026 · How we rate
Sources
- 1.Louvre jewel heist: fake AI videos of the robbery circulate online, France 24 (Truth or Fake) (2025)
- 2.AI-generated clip of Louvre jewel heist spreads online, AFP (via Yahoo News) (2025)
- 3.Fact-Check: AI-generated scenes ‘from France’s 2025 Louvre heist’ circulate in Hong Kong, South China Morning Post (2025)
- 4.AI-generated clip of Louvre jewel heist spreads online, CEDMO (European Digital Media Observatory hub) (2025)
- 5.2025 Louvre heist, Wikipedia
- 6.Four more suspects detained over Louvre jewel heist, CNN (2025)
- 7.Louvre’s Apollo Gallery reopens after French Crown Jewels theft scandal, Euronews (2026)
Help us investigate
This is a living case file. If you spot an error or know evidence we missed, tell us, and weigh in on where you land.
Where do you land?
Cast your read on this one.
Comments
Add your take. Comments are read and approved by a human before they appear, so keep it on topic and civil. Please do not accuse named, living people of crimes.