When Images Stop Being Evidence: What a Small Set of Recent Deepfake and Manipulated Photo Incidents Tells Us

Article

In this sample of incidents we’ve reviewed here at Realz, the cases come from only a handful of countries — Belgium, Germany, Lebanon, and Sweden — over a short period from early February to early April 2026. That is far too small and uneven a set to support sweeping conclusions about the whole information environment.

Even so, the incidents reviewed here point to something important: image deception is not one problem. It is several related problems that all increase the burden of verification.

Some of the reported cases involve AI-generated sexualized content targeting real people. Others involve impersonation through fake social media accounts and synthetic visuals. Another involves an officially circulated manipulated image tied to a deadly real-world event. In each case, the image itself matters, but the larger issue is what the image is used to make people believe.

A narrow sample, but a useful one

Across this small set of reported cases, three patterns stand out.

First, public figures remain obvious targets. In Belgium, multiple outlets reported that false AI-generated images and videos of Crown Princess Elisabeth circulated through a fake Facebook profile posing as the princess. The content reportedly included dozens of videos and images, some sexualized or otherwise designed to attract attention, and some users reportedly did not understand that the material was AI-generated. The Belgian royal household said it was aware of the false content and was taking steps to get it removed.

Second, non-consensual synthetic imagery continues to blur the line between harassment, impersonation, and reputational harm. Reporting around German actor and media personality Collien Fernandes describes allegations that false nude images, deepfake pornography, and fake online identities were used in her name over a period of years. Important facts in that case remain contested: Fernandes has publicly accused her ex-husband, Christian Ulmen, while Ulmen has denied spreading deepfake content. The reporting nevertheless shows how synthetic or allegedly synthetic imagery can become part of a much broader pattern of identity abuse.

Third, manipulated visuals can be used not just to embarrass or exploit, but to support highly consequential claims. In Lebanon, reporting said that after three journalists were killed in an Israeli attack, Israel’s military circulated an image presenting one of the dead journalists, Ali Shoeib, in military uniform. According to the reporting, the military later acknowledged that no such authentic image existed and that the image had been photoshopped. That matters because the manipulation was tied to a claim about the identity and role of a person who had just been killed.

These are very different incidents. But they all show the same structural issue: once a false or manipulated image enters a public channel, the burden shifts to everyone else to prove what is real.

Deepfakes are only part of the story

One mistake in public discussion is to treat every false image as the same kind of phenomenon. The incidents we’ve looked at here at Realz suggest a more useful distinction.

Some cases are reported as deepfakes or AI-generated imagery. The Belgium incidents involving Princess Elisabeth, and the Sweden case involving a false image of Magdalena Andersson with Jeffrey Epstein, fall into that category based on the reporting. In the Swedish case, the image was reportedly shared in a Facebook group with more than 150,000 members, and discussion in the comments showed that some users considered it plausible while others questioned it. Andersson’s press secretary said the image was fake and that she had never met Epstein.

Other cases are better described as manipulated photos or visual impersonation. The Lebanon incident is the clearest example in this sample. The issue there was not simply that an image was fake, but that an altered image was used to reinforce a serious public allegation in the aftermath of a lethal attack.

That distinction matters because trust failures do not depend on technical sophistication. A convincingly edited image, a synthetic portrait, or a fake account publishing fabricated visuals can all do damage if they fit a story people are already inclined to believe.

The real impact is on trust and decision quality

In a narrow sense, these incidents are about pictures. In a broader sense, they are about decision-making.

When sexualized fake images of a public figure spread through social platforms, the immediate harm may be humiliation, reputational damage, or loss of control over one’s likeness. When a false image links a political leader to a notorious figure, the image is trying to influence public judgment, even if the fabrication is later exposed. When an official actor distributes a manipulated image connected to a journalist killed in war, the stakes move beyond embarrassment into narrative shaping around legitimacy, responsibility, and evidence.

That is why image authenticity should not be treated as only a media-literacy issue. It is also a governance issue.

Organizations, institutions, and public-facing teams increasingly have to answer a basic question quickly: what confidence do we have that this image is genuine, unaltered, and correctly presented?

If they cannot answer that question with a disciplined process, they risk making poor decisions, amplifying falsehoods, or responding too slowly while manipulated content spreads.

Social platforms remain an obvious amplifier

Several incidents in this small sample involve distribution through social media rather than obscure technical channels.

The Belgium reporting points to a fake Facebook profile with nearly 30,000 followers. The Swedish case also centered on Facebook group distribution. The reported harm in both cases depended less on advanced tradecraft than on scale, visibility, and plausibility.

This is an important reminder: many visual-authenticity incidents do not begin with elite technical deception. They begin with ordinary platform mechanics — reposting, impersonation accounts, engagement incentives, and fragmented moderation.

That does not make them minor. It makes them easier to spread.

Verification is getting harder, not impossible

It would be easy to overreact to incidents like these and say that images can no longer be trusted at all. That would be too broad, and not especially helpful.

A better conclusion from the incidents reviewed here is more specific: visual evidence now carries more uncertainty than many people and institutions are used to handling.

The Swedish example is instructive. According to the reporting, some viewers accepted the fabricated image of Magdalena Andersson and Jeffrey Epstein as potentially real, while others noticed irregularities and questioned it. That split reaction is part of the problem. False images do not need to convince everyone. They only need to create enough uncertainty, outrage, or suspicion to influence conversation.

Likewise, in the Belgium incidents, reporting said some users did not understand that the videos were AI-generated. That shows how synthetic visuals can exploit not just technical gaps, but contextual ones: people often encounter images without source history, without provenance, and without any reason to pause before reacting.

What leaders should take from this

In the incidents we’ve looked at here at Realz, the recurring lesson is not that every organization needs perfect deepfake detection. It is that public trust now depends on better verification habits.

A few practical implications follow from that:

1. Treat image authenticity as an operational question

If an image could affect reputation, legal position, public communications, or executive decision-making, it should not be handled casually. Teams need a clear path for escalation and verification.

2. Distinguish between allegation, reporting, and confirmation

Several incidents in this sample remain reported but not fully verified, or involve disputed claims. That uncertainty should be preserved in public communication. Overstating confidence can create a second trust failure.

3. Verify source, context, and provenance — not just pixels

A convincing image can still be false in origin, false in caption, or false in implication. The key questions are who posted it, where it first appeared, whether the depicted event independently happened, and whether the affected party or institution has confirmed or denied it.

4. Prepare for impersonation at the account level

The Belgium cases underline that false visuals often travel through false identities. A fake profile that appears plausible can be just as important as the media itself.

5. Assume reputational harm can arrive before verification does

By the time a palace, press office, or legal team responds, an image may already have circulated widely. Response planning matters because image incidents move at platform speed, while verification often moves much more slowly.

The bigger issue is not the image alone

In this small editorial sample, the most important pattern is not simply that fake images exist. We already knew that. The more useful observation is that manipulated and synthetic visuals are showing up in different settings with different kinds of harm: harassment, identity abuse, public confusion, and narrative manipulation.

That is why the right response is not panic. It is discipline.

Visual authenticity now requires stronger habits around verification, clearer institutional response, and more humility about what can be concluded from a single image seen on a social feed. In the incidents reviewed here, the people and institutions under pressure were often forced into a reactive position, trying to disprove content after it had already reached an audience.

That is the trust problem in its clearest form.

An image no longer needs to be real to have consequences. It only needs to be believable long enough.

— Ken Andersen