Fact-checks

Julian Assange in prison: the SI image made by a supporter

A tilted cream instant photo showing a pillow and a stained blanket, next to a barred window, under an amber glow; the right edge of the photo breaks up into amber pixels that drift away
Illustration created with Claude (Anthropic).

In March 2023, a picture of a frail Julian Assange slumped under a blanket in prison went viral as his latest photo. It was SI-generated by one of his supporters. Here is how fact-checkers exposed it and what our detector returns on the original post's file.

A man with long white hair lies slumped against a pillow, eyes closed, face swollen and red, a stained cream blanket pulled up to his chin. Behind him, a white tiled wall. Across the picture, a diagonal watermark reads "Photo property of 'E'". At the end of March 2023, this image went around the world as the latest photo of Julian Assange, the WikiLeaks founder then held in Belmarsh prison in London.

It was not a photo. The picture of Assange in prison is an AI image, or in our words an SI-generated one, typed into Midjourney by one of his own supporters. Fact-checkers in Britain, the United States, Spain, Germany and India took it apart within days or weeks, and its author admitted it openly. This article sums up what they found, why a fake made by a supporter is a case of its own, and what our check for SI tool returns on the file attached to the original post.

Claim: a recent photo shows Julian Assange weak and exhausted in a British prison.
Verdict: false. The image was made with the SI image generator Midjourney by a Twitter user who signs as "E" and campaigns for Assange's release. He confirmed it to several newsrooms, and Assange's wife told the German newspaper Bild that the picture was fake, according to CORRECTIV. Full Fact rated it altered, BOOM rated it false and CORRECTIV rated it manipulated. Our detector scored the posted file 99% likely SI.

Where did the picture come from?

The trail is unusually short. Every fact-check that traced the image ended at the same account, The Errant Friend, whose profile read "Free Assange". The post is timestamped 01:18 UTC on 30 March 2023, which explains why Lead Stories dates it to 29 March while BOOM, Full Fact, The Quint and CORRECTIV give 30 March: it was still the evening of the 29th in the Americas. CORRECTIV called it the earliest source it could find.

The post did not pretend to be a leak. It read: "You absolutely MUST free Julian Assange. It is time sensitive, and urgent. He does not have another year to wait behind bars while the wheels of "justice" crush him slowly." The picture underneath carried the watermark that later led fact-checkers back to it.

It was not the account's first attempt. On 17 March 2023, the same user had posted another generated image of Assange behind bars, with the caption "How far would you go, to free Julian Assange?". The Quint and Lead Stories noted that both pictures carried the "E" watermark. Lead Stories added that the account had also posted generated images of Donald Trump, one of which Trump himself shared on Truth Social. The user described himself to Lead Stories as a "former central developer of the AssangeDAO and other Assange related projects".

How did the image spread?

Within days, the watermark stopped meaning anything to the people sharing the picture. On Facebook, BOOM found it posted as the "Latest picture of Julian Assange in the UK prison." Maldita, the Spanish fact-checker, collected replies such as "This is pure evil of what the US and UK are doing to this poor man!" and claims that the photo had been leaked from his isolation cell. CORRECTIV saw Facebook posts calling it his "last photo", and wrote that the image was being shown hundreds of thousands of times.

It also travelled through channels with large audiences. CORRECTIV listed popular Telegram channels, among them one run by a Ukrainian politician, and posts on the Russian social network VKontakte claiming Assange was being "slowly killed" (our translation). The official Twitter account of the Russian Embassy in Kenya shared it too, as BOOM, The Quint and CORRECTIV all reported, with a caption that turned the picture into a diplomatic message: "13 years ago, Julian Assange released a video showing US Forces firing on Iraqi civilians, including children for no reason. Now he faces up to 175 years in prison for publishing it."

Made by an activist to move his own followers, the image ended up in a state account's argument against the United States, without a word about how it had been made.

How fact-checkers proved it was fake

In most of our fact-checks, the case rests on clues: no photographer, a wrong setting, deformed hands. Here the fact-checkers had something rarer: a confession, backed by evidence.

The creator said so, again and again

Bild interviewed the account owner first. Full Fact quoted his explanation to the German tabloid: "My intention was to create an image based on the documented happenings around Julian." He told Lead Stories, by direct message on 5 April 2023, that he had used Midjourney: "I used MidJourneys version 5, an AI image creator to carefully manufacture it, and then watermarked the photo for effect and so people understood where it came from." He confirmed the same to BOOM, calling the picture a "viral art project", and to The Quint.

He also offered proof. BOOM says he shared a screenshot of the prompt he typed. CORRECTIV reports that he confirmed his authorship to AFP and Reuters, and that Reuters published his screenshot of Midjourney showing the text command and the resulting image. In CORRECTIV's German rendering, which we translate back, the prompt asked for a photo of Assange held in Belmarsh, starved, emaciated and broken but determined to fight the proceedings, slumped on a dirty bench in the infirmary, still alive, in a gloomy atmosphere. That description matches the picture closely.

His wife and his allies did not share it

The second line of evidence came from Assange's own side. Full Fact quoted Guillaume Brossard, co-founder of the French site Hoaxbuster, who pointed out that Assange's close allies, including his wife, were active on Twitter but had not shared the image. CORRECTIV and The Quint both report that Stella Assange told Bild the picture was a fake.

That is a useful reflex: the people closest to a famous prisoner have every reason to publish real evidence of mistreatment. Their silence is not proof, but it is a strong hint.

The details did not hold up

Then came the pixels. Brossard noted, in Full Fact's account, that the hair, the ear and the mismatched sleeve colours did not look realistic. CORRECTIV spotted an odd colour gradient on the blanket at Assange's shoulder, eyelids that look smooth and blurred, and a mark on the side of his nose that looks like half a piercing. Maldita pointed at the fuzzy texture of the hair and skin and at a very pronounced shadow under the cheekbone.

None of these flaws would settle the case alone: real photos taken in poor light can look smeared too. Combined with the watermark, the confession and the prompt, they leave no doubt.

A fake made by a supporter

Most SI fakes we have covered were made as jokes, as bait for clicks, or to embarrass someone. This one was made by a sympathiser, for a cause he believed in, and he did not hide it. In a statement on his Telegram channel quoted by The Quint, he wrote: "I sought to make his documented suffering real, and succeeded in doing so." CORRECTIV quoted another of his posts there, in which he said the photo did not need to be real to depict reality (our translation).

He told Lead Stories that Assange had been "basically barred from public appearance" since his arrest, and that the image was "designed to evoke a visceral response and to accurately represent what the public could not otherwise bear witness to."

Our view: whatever one thinks of Assange's case, a generated picture cannot represent what nobody saw. It can only show what its maker imagined. Shared without its context, it becomes a false document about a real person's health, at a time when his lawyers and family were making their own claims through real channels. It also hands an easy argument to anyone who wants to dismiss genuine evidence later. The watermark was meant to sign the work. In practice, almost nobody who shared it read it as a warning.

The same month, two other Midjourney pictures spread widely: the arrest of Donald Trump, told in our fact-check of the Trump arrest images, and the Pope in a puffer jacket, in our fact-check of the Pope puffer jacket image. Full Fact drew the link at the time. Like the picture of Putin kneeling before Xi Jinping, the Assange image targeted a famous man at a moment when the world was already watching him.

Why did it come back in 2024?

Fakes rarely die after a fact-check. In March 2024, SEE Check republished in English a fact-check by Faktoje.al, from Albania. The same image had been circulating on Albanian-language social media since February 2024, this time with the claim that Assange had been assaulted in jail. Users compared it with the death of the Russian opposition leader Alexei Navalny to accuse Western media of hypocrisy.

A reverse image search on TinEye showed Faktoje that the picture had been online since at least April 2023, nearly a year before the new caption. The same fact-check notes that the "Don't Extradite Assange" committee had told AFP on 4 April 2023 that the image was made with SI. Our guide on how to reverse image search explains how to run that check, which was enough to date the 2024 version.

We tested the image in our SI detector

For once, a copy close to the source exists: the file attached to The Errant Friend's post of 30 March 2023, a JPEG of 877 by 1,023 pixels with the "E" watermark. It is not the file that came out of Midjourney, since Twitter usually re-encodes uploaded images, but it is the version the first viewers saw, before any screenshot or repost. Because it has no free licence and shows a real person in a fabricated state, we describe it here rather than reproduce it.

We ran it once through our SI image detector, exactly as a visitor would.

Result: 99% likely SI. The classifier found the statistical patterns of SI-generated images in the pixels (classifier: SI 99%). The face fills a large part of the frame, so the site also checked the face close-up in detail before deciding.

Metadata: JPEG, 877 by 1,023 pixels, about 159 KB. No camera make, model or date, a weak hint on its own. No Content Credentials and no XMP tag declaring SI generation.

What the test covers: this one copy, as attached to the original post. It does not identify the generator (we know it was Midjourney only because its author said so), and it says nothing about the reposted versions that circulated in 2024.

The verdict is as clear as on other files close to their source, and clearer than the 82% we got on a cropped screenshot in the Putin case. That fits what we see again and again: every screenshot, crop and recompression erodes the signal a classifier reads. The empty metadata proves nothing either way, since social networks strip it from real photos too. A score, however high, is one clue among others, as our article on how accurate SI detectors are explains. In this case, the confession and the prompt were stronger evidence than any detector.

What happened to Julian Assange?

The real story ended a little over a year later, with real photographs. According to NBC News, Assange pleaded guilty in June 2024, in a US federal court in Saipan in the Northern Mariana Islands, to a single count of conspiracy to obtain and disclose national defense information. He was sentenced to the 62 months he had already served in prison in Britain, and flew home to Canberra, Australia, where he hugged his wife after landing. NBC reports that he had spent more than five years in Belmarsh fighting extradition to the United States.

The pictures of that arrival were taken by press photographers and carried agency credits. That is exactly what the March 2023 image never had.

What this case teaches

  1. Read the watermark. A name stamped across a "leaked" photo is a lead. Searching for it took fact-checkers straight to the author.
  2. Ask how the photo could exist. Pictures from inside a high-security prison infirmary do not usually reach social media. If nobody can explain who took it, be careful.
  3. Check the people closest to the subject. Family, lawyers and campaign groups publish real evidence when they have it. Their silence, or their denial, matters.
  4. Look at skin, hair and fabric. Smooth, blurred eyelids, smeared hair and odd colour shifts in cloth were the visible flaws here.
  5. Distrust the picture that fits your cause. This fake was made by a supporter and shared by people who agreed with him. Good intentions do not make an image true.

Our guide on how to spot SI images walks through these visual checks step by step, and the SI or Not game lets you test your eye on real and generated pictures.

FAQ

Is the photo of Julian Assange in prison real?

No. The viral picture of a weak, exhausted Julian Assange under a stained blanket was generated with Midjourney by a Twitter user called The Errant Friend, who signs as E and campaigns for Assange's release. He confirmed it to Bild, Lead Stories, BOOM and The Quint, and Full Fact, BOOM, Maldita and CORRECTIV all concluded it was not a real photo.

Who made the AI image of Julian Assange?

A supporter of Assange using the Twitter account The Errant Friend, with the watermark Photo property of E. He told Lead Stories he used Midjourney version 5 and added the watermark so people would know where the image came from. He described his goal as showing what the public could not see.

How can you tell the Assange prison picture is AI-generated?

Its author admitted it and showed his Midjourney prompt. Assange's allies, including his wife, did not share it, and Stella Assange told Bild it was fake. Fact-checkers also pointed to unrealistic hair and skin, smooth blurred eyelids, an odd colour shift in the blanket and mismatched sleeve colours. Our SI detector scored the file from the original post 99% likely SI.

Did the Russian embassy share the fake Assange photo?

Yes. BOOM, The Quint and CORRECTIV reported that the official Twitter account of the Russian Embassy in Kenya posted the image with a caption about Assange facing up to 175 years in prison, without saying it was generated.

Is Julian Assange still in prison?

No. According to NBC News, he pleaded guilty in June 2024 in a US federal court in Saipan to a single count of conspiracy to obtain and disclose national defense information, was sentenced to the 62 months he had already served in Britain, and returned to Australia.

Sources

  1. 'Recent photo' of Julian Assange was generated by AI · Full Fact, 3 April 2023
  2. Fact Check: Social Media Image NOT Photo Of WikiLeaks Founder Julian Assange -- It's AI-Generated Deepfake · Lead Stories, 3 April 2023
  3. No, esta imagen de Julian Assange en la cárcel no es real: está generada con inteligencia artificial · Maldita.es, 3 April 2023
  4. Image of Julian Assange in Prison Is Not Real But AI-Generated! · The Quint, 11 April 2023
  5. AI Generated Photo Of Julian Assange In Prison Goes Viral · BOOM, 12 April 2023
  6. Künstliche Intelligenz: Nein, diese Aufnahme von Julian Assange im Gefängnis ist nicht echt · CORRECTIV, 20 April 2023
  7. Viral Image of Julian Assange Mistreated in Jail Actually Created by AI · SEE Check (Faktoje.al), 25 March 2024
  8. Julian Assange returns home to Australia a free man after U.S. plea deal · NBC News, 26 June 2024

Spotted an error? Write to hello@siornot.com. Corrections are made in the article and the update date changes. Read our editorial policy.