This fake Mads Mikkelsen picture proves AI is getting scarily realistic

Recently, the internet was ablaze with envy after a woman bumped into actor Mads Mikkelsen while celebrating her birthday. Sharing the celebrity snap online, she noted that Mads was celebrating his 60th – a coincidence that sounded too good to be true. Well, that's because it was, and AI is to blame.

Whether we like it or not, AI slop is all over the internet, and s**t's getting real. Gone are the days of extra fingers and uncanny faces; today's AI creations are getting more refined by the second, fooling social media with worrying ease. While there are undoubtedly ways to use AI properly, it's hard not to feel a little existential.

Posted on X by software developer Aya Bochman, the innocent snap of her and Mikkelsen was a viral hit, as fans flocked to the post with birthday wishes to the pair. Matching the lighting, texture and perspective of the original image, the AI addition was practically undetectable. It seemed the internet was fooled... mostly.

A few eagle-eyed viewers felt like something fishy was going on, and after some digging, it was discovered that 25-50% of the image presented a SynthID watermark. Busted. But why did Aya do it?

In a new tweet, she explained that the post was merely meant to be an inside joke for close friends and followers. "For those who follow me, I thought it would be obvious. But as soon as it reached the rest of the world...oh boy," she wrote. As the co-founder of FASHN, an AI studio for fashion brands, Aya's post should've rung some alarm bells, but the innocuous image went unscrutinised for some time.

It goes to show how convincing AI images have become, thanks to the likes of Sora 2 and Google V3. The technology is more refined and more accessible than ever, making it harder to distinguish reality from prompted content. While Aya's playful prank shows the lighter side of AI, it raises concerns for the future of social media and the spread of misinformation. Never has there been a more important time for proper AI regulation; labels and community notes don't cut it anymore.