Even Katy Perry's mom was fooled by what appeared to be AI-generated Met Gala pics

Several celebrities who couldn’t make it to the Met Gala showed up online anyway — in fake images that showed signs of having been generated by artificial intelligence.

The fake images, which depicted them posing in glamorous garden-themed ensembles to match this year’s dress code, were quickly shared thousands of times on X and Facebook amid the flurry of real photos flooding social media of celebrities walking the carpet Monday evening.

Two photos of pop star Katy Perry that appeared to have been created by generative AI accrued hundreds of thousands of engagements after they were shared on X. One depicted her in a floor-length cream gown adorned with flowers and lined with moss, while the other had her wearing a metallic corset and a leafy skirt of flowers and vines. The picture with the cream gown appeared online as early as April 18.

But Perry wasn’t there; she later posted on Instagram that she was tied up in the recording studio, instead. The post included a screenshot of a text exchange showing that even her mother fell for the fake photos.

“lol mom the AI got you too, BEWARE!” she texted in response.

Fake images also circulated of the singer Rihanna posing at the Met Gala in a flashy gown even though she skipped this year’s event because she was reportedly ill with the flu. Similar images of Billie Eilish, who also wasn’t there Monday, were shared online, as well.

The Metropolitan Museum of Art, which hosts the Met Gala, didn’t immediately respond to a request for comment.

The pictures are the latest incident of fake, seemingly AI-generated content fooling internet users en masse, from images that ultimately become memes (like that of Pope Francis in a Balenciaga puffer coat) to ones for political purposes or manipulate public opinion (such as an AI-generated image of former President Donald Trump posing with Black voters). Facebook has a bustling community of pages that share bizarre AI images of everything from Shrimp Jesus with flight attendants to extremely muscular cows.

Those who look closely at Perry’s viral photo may be able to find some telltale signs of AI creation — many pointed out a nonsensical arrangement of photographers in the background, as well as inconsistencies in the design of the carpet — but the images appear photorealistic at first glance.

Some experts said it’s even easier to fall for fake images when they’re placed in plausible contexts. People are more inclined to believe that celebrities wore outlandish outfits to the Met Gala because that’s what the event is designed for, and they have attended in similarly glam outfits in the past, whereas outright political disinformation may be more likely to earn scrutiny.

Yotam Ophir, an assistant professor of communication at the University at Buffalo studying misinformation, said that while this type of fake photo seems harmless on the surface, social media algorithms offer incentives to create them because fake images can easily generate revenue-earning clicks and reactions for the users who post them.

“It almost baits people to catch them lying, like it’s intended for people to say, ‘Hey, that’s AI,’” Ophir said. “But AI is a hot topic right now ... so they’re going to talk about it. And the fact that Katy Perry’s mom fell for it just made it even more viral. That’s quite an achievement for somebody who produced misinformation probably for free or almost free.”

Major social platforms such as X, Meta and TikTok have tried to regulate the proliferation of AI misinformation by requiring AI-generated media to be clearly labeled, but no platform has yet implemented a consistent mechanism able to identify all such content.

As AI creation tools grow more sophisticated over time and it becomes easier for everyday users to disseminate misinformation, experts said, the internet may soon enter an era when people are unwilling to trust anything they see online.

“Right now we are in the stage where we still automatically believe what we see. It’s an evolutionary mechanism to trust our eyes,” Ophir said. “But we’re not wired to cope with such a massive and industrialized version of deception as what we have now coming from AI.”

While some, like Ophir, are alarmed about what that may mean for public discourse, other scholars are more optimistic. Subbarao Kambhampati, a professor of computer science at Arizona State University researching AI, said he believes the current disinformation ecosystem will teach users to be more discerning with their news and content consumption.

People might be forced to make it a habit to check the sources of what they see online, he said. Meanwhile, improved cryptographic authentication will also make it easier to technologically identify fake media.

He said the average internet user is most likely not yet well-informed about what AI models can do and how to spot their output, so people are much more easily tricked by the technologies now. But he said he thinks people will quickly learn how to navigate a “new normal” in which everything looks believable, real or not.

“Even Katy Perry’s mom probably won’t fall for [the fake Perry image] next year,” Kambhampati said. “The technology changes the landscape, but the interesting thing is we are amazingly good at adapting.”