The leading deepfake expert has given up: it is no longer possible to tell fake from reality

A few years ago, a fake video could be spotted by its odd facial expressions, unnatural lip movements or “floating” shadows. Today the situation has changed so much that alarm is being sounded not by journalists and politicians, but by people who have devoted their entire careers to fighting fakes.

One of the world’s leading digital forensics specialists, Hany Farid, admitted that modern deepfakes have reached a level where even he finds it increasingly difficult to distinguish fiction from reality. For more than twenty years, the scientist has been exposing fake photos, videos and audio recordings, advising governments, law enforcement agencies and the world’s largest media outlets. But now, he says, the usual verification methods are beginning to fail.

Farid says he can no longer simply look at an image and understand whether it is real or not. Every photo has to be investigated almost like a case file: analyzing the lighting, checking the geometry of shadows, studying reflections, and assessing the movement of objects frame by frame. Even then, this approach does not always guarantee the right result. According to the scientist, people today identify AI-generated content only slightly better than random guessing.

The most troubling thing is not the quality of the fakes, but how accessible their creation has become. If producing a convincing fake once required significant computing power and specialized knowledge, now all it takes is a laptop and a few minutes of free time. Tools that until recently were available only to states and large organizations are now openly accessible to billions of people around the world.

The consequences are already becoming visible. Scammers are actively using cloned voices of relatives to extort money, creating fake appeals from celebrities and politicians, and also faking video calls and voice messages. In recent weeks, deepfakes of well-known British politicians and officials have been circulating online and used to promote financial scams.

But an even more dangerous effect is another one, which researchers call the “liar’s dividend.” If fakes once created the risk of disinformation, now the opposite problem is emerging: any genuine recording can be dismissed as fake. A video of a crime, a corruption scandal or a war crime can be rejected simply because the public is no longer sure that objective visual reality exists.

In Farid’s view, humanity is approaching the point where photographs and video recordings will no longer be considered stand-alone evidence. Instead of trusting our eyes, we will have to rely on complex systems for verifying file provenance, digital signatures and methods of confirming content authenticity from the moment it is created.

“I don’t trust anything anymore,” the expert admitted in one recent interview. And if a person who has spent decades exposing digital forgeries says such things, then perhaps the era when a photograph was considered irrefutable proof is indeed coming to an end.