Trends

The Photos Nobody Took: What Unlabeled AI Images Are Doing to the Internet

For most of the history of photography, a photo was a kind of proof. Someone was there, pointed a camera and pressed a button. Photos could be staged or edited, but making a convincing fake took skill and time.

That has changed. Today anyone can make a realistic image of a place, a person or an event in seconds. Many people use this for art, design and fun, and they say so. This article is not about them. It is about images that are made with AI and then shared as if someone had taken them. We call them the photos nobody took.

A photo used to be evidence

People are not good at spotting AI images by eye. In a 2022 study published in PNAS, Sophie Nightingale and Hany Farid asked 315 people to sort faces into real and AI generated. Their accuracy was 48 percent, about the same as guessing. The same study found that people rated the AI faces as slightly more trustworthy than the real ones.

Generators have improved a lot since then. So the old advice, "look for strange hands", helps less every year. When we cannot trust our eyes, we start to doubt every photo. That doubt is the real cost.

It also lands on news. The Reuters Institute's Digital News Report 2025 found that overall trust in news stayed at 40 percent across the countries it studied, the same as the two years before. Images that look like news photos but were never taken make that harder to fix.

When news breaks

Fake images spread fastest when people are scared and want to help. After Hurricane Helene in 2024, AI images of flooded towns and rescued children went viral. Fact checkers at Lead Stories traced one widely shared image of a girl holding a puppy in a boat and found no news source for it. Another version of the same scene showed a different shirt, a different boat and an extra finger.

Experts at Virginia Tech warned at the time that such images can erode trust in real news during a crisis, and can also be used to lure people into scams. A fake picture of a disaster can pull attention and donations away from the people who really need them.

Most people who shared these images meant well. That is the point. Good intentions spread a fake just as fast as bad ones.

Scams, products and reviews

Generated images also make it cheaper to look real in commerce. A product that does not exist can have perfect photos. A seller can show "customer photos" that no customer took. A fake profile can have a friendly face that belongs to nobody.

Regulators have started to respond. In August 2024 the US Federal Trade Commission finalized a rule against fake reviews and testimonials, and said it also covers reviews made with AI. The rule took effect in October 2024. It does not end fake reviews, but it makes clear that "an AI wrote it" is not an excuse.

For shoppers the lesson is simple. Photos in a listing or a review are a claim, not proof. Look for reviews with ordinary, imperfect photos, check the seller's history, and be careful when every picture looks like an ad.

The liar's dividend

There is a second, quieter harm. In 2018 the law professors Bobby Chesney and Danielle Citron described what they called the liar's dividend. When everyone knows that fakes are easy to make, a person caught on a real photo or video can simply say "that is fake", and some people will believe it.

This has already reached courtrooms. NPR reported in 2023 that people had started to claim that real videos used as evidence were deepfakes, and that judges were so far not accepting the argument without proof. Still, the more fakes there are, the easier it becomes to wave away something true.

So unlabeled AI images do not only add false pictures to the world. They also make true pictures easier to deny.

What labels can and cannot do

The most common answer is labeling. Some AI tools and some cameras now attach Content Credentials, information under an open standard called C2PA that can record how a file was made and edited. Some platforms say they add a label when they detect that an image was made with AI.

Laws are moving too. In the EU, the AI Act requires clear labels on deepfakes from 2 August 2026. That covers AI images or videos that closely resemble real people, places or events and could be taken as authentic. Clearly artistic, satirical or fictional work gets a lighter duty, so a label does not have to spoil it. How this works in practice is still being settled.

Labels have limits. Content Credentials are easy to lose: a screenshot, a crop or a re-upload often removes them. Detection on platforms misses some images and flags some real ones. And a missing label proves nothing. Most real photos have no label either.

So labels are useful when they are there. They cannot be the only defence.

What you can do

• Pause before you share. Strong feelings are exactly when fakes spread.
• Ask where it came from. Who took it, and where was it first posted? A photo with no source is a rumour.
• Search for it. A reverse image search often finds the original, or a fact check.
• Label your own AI images. If you make AI art, say so. It costs nothing and keeps the line clear.
• Do not call a real photo fake without a reason. Wrongly calling something fake is its own kind of harm.

For a step by step method, see how to tell if a photo is real or AI-generated.

What we did at Rawly

Rawly is a photo and video challenge app. We did not try to build a better AI detector. We tried to make the question easier to answer inside our own app.

• Live camera only. Every photo and video is taken with the camera inside the app. There is no gallery upload, so there is no file to upload a generated image from.
• A code you can check. Every photo shared from Rawly carries a short code. Anyone can type it at rawly.app/verify and see Rawly's copy of the photo and the minute it was taken.

We want to be honest about the limits. The check proves when a photo was taken in Rawly, not who took it. Someone could still point the camera at a screen, and we rely on reports for that. And the check says nothing about photos from other apps. If you want more apps built this way, see AI-free social media apps.

Questions

What is an unlabeled AI image?

It is an image made or heavily changed with AI and shared as if it were a real photo, with no label or note that says how it was made. The image itself is not the problem. The missing context is.

Is it wrong to make AI images?

No. AI images can be art, design, satire or fun, and many people label them clearly. The harm starts when an AI image is presented as a real photo of a real event, place, person or product.

What is the liar's dividend?

It is a term from law professors Bobby Chesney and Danielle Citron. When people know that fakes are easy to make, someone caught on a real photo or video can claim it is fake. Doubt itself becomes useful to the person who wants to deny the truth.

Are there rules about labeling AI images?

Some, and they are new. In the EU, the AI Act requires clear labels on deepfakes, including AI images of real looking people, places or events, from 2 August 2026. In the US, the FTC's rule against fake reviews, in force since October 2024, also covers reviews made with AI. Rules differ by country, and this is not legal advice.

How can I avoid sharing a fake photo?

Pause before you share. Ask who took it and where it was first posted, run a reverse image search, and look for a news or fact check source. If you cannot find where it came from, do not share it as fact.

How does Rawly help?

Rawly only accepts photos and videos taken live with the camera inside the app. Every photo shared from Rawly carries a short code, and anyone can type it at rawly.app/verify to see Rawly's copy and the minute it was taken. It proves when a photo was taken in Rawly, not who took it, and it says nothing about photos from other apps.