Some of the biggest names in music, entertainment and fashion assembled in New York City for Monday’s “Garden of Time”-themed Met Gala, decked out in flowers, sparkles and extravagant timepieces.
Contrary to the images circulating on social media, Katy Perry was not one of them.
“Couldn’t make it to the MET, had to work,” the singer posted on Instagram, alongside a video of herself singing in the studio — as well as two photos seemingly showing her at the gala.
They were actually made with AI.
In the first, Perry appears to be standing on a hedge-lined red carpet at the Met, wearing an elaborate ball gown covered in flowers and butterflies, with her dark hair styled in long waves. In the second, a close-up shot, Perry is wearing a metallic corset top with a large key handle down the middle and a short skirt of flowers and leaves, her hair straight and tousled.
The photos — whose exact origin is unclear — made a splash on X (formerly known as Twitter) earlier in the night, as viewers at home refreshed their feeds and weighed in on their favorite celebrity fits.
One X post of the ball gown photo had over 300,000 likes and nearly 70,000 reposts as of Tuesday morning — and a note at the bottom clarifying that it was created with AI. Another post, of the corset outfit, garnered over 100,000 likes and was eventually labeled “digitally created.”
Perry is a regular Met Gala attendee (and famous for dressing on-theme, including as a chandelier and hamburger in recent years), so internet observers would be forgiven for assuming she was there on Monday.
Even Perry’s mom, Mary Hudson, thought so. One of the posts in the singer’s Instagram carousel was a screenshot of a text conversation between the two.
“Didn’t know you went to the Met,” her mom wrote. “What a gorgeous gown, you look like the Rose Parade, you are your own float lol.”
Perry was quick to clear things up.
“Lol mom the AI got you too,” she replied. “BEWARE!”
AI-generated images are increasingly easy to make, and celebrity deep fakes are increasingly prevalent — from sexually explicit deep fakes of Taylor Swift circulating earlier this year to robocalls imitating President Biden ahead of the New Hampshire primary.
In fact, Perry wasn’t the only star to be basically photoshopped onto the Met Gala red carpet: A viral X post claimed to show an elaborately dressed Rihanna in attendance, when she was actually home sick with the flu.
Another purported to show Lady Gaga, who hasn’t been there since 2019.
And before the gala, photos circulated of Dua Lipa wearing bangs and a corset, only for her to show up on the carpet with crimson hair and an all-black ensemble — and for an X user to point out the early photos were from a 2021 Vogue shoot.
While some social media users dismissed the Met Gala fakes as part of the fun, others see them as a worrisome sign of what could lie ahead.
Experts warn AI-generated deep fakes pose a threat to everything from election security to everyday scams. And just last month, more than 200 artists — including Perry herself — signed a letter urging tech platforms and digital music services to stop using AI to “infringe upon and devalue the rights of human artists.”
So what clues should Mary Hudson — and other discerning viewers — be looking for to spot potential fakes?
Sam Gregory of the nonprofit Witness, which helps people use video and technology to protect human rights, encourages viewers to rely on context and intuition in situations like this one.
“My starting point with all images like [Perry’s] is to not trust the online detectors as there are too many variables around whether they give an accurate result,” he explained over email.
He said when he ran both Perry images through a widely-used detector, the flower dress came back as “likely human” and the corset as “likely AI generated.” He also discourages people from looking for visual clues in these kinds of images, saying that can “lead down a rabbit hole of unproductive forensic skepticism.”
With a high-profile event like the Met Gala, Gregory says, it’s best to use “classic media literacy and verification approaches.” In this case, that could mean looking for more proof of Perry’s attendance, from a variety of sources.
“Although some media literacy strategies like checking the source might lead us astray — perhaps we do trust Katy Perry to share real images of herself — if we use another strategy and look for other images from the same event from reliable sources, we’d quickly see this isn’t real,” he explained.
He adds that it reminds him of the AI-created images of a fire at the Pentagon that went viral last year. In both cases, he says, the first question people should ask themselves is not whether they can spot the AI glitches in the photo, but “Why aren’t there other photos and videos of this event in a highly populated area?”
But the bigger question, as Gregory sees it, is whether the public should be expected to do this at all.
“Why wouldn’t Katy Perry be at the Met Gala and why would we second-guess that, particularly if she’s part of the deception?” he says.
More help may be coming from social media platforms, amid growing concerns about the potential for AI to mislead users. Meta said earlier this year that it would start labeling AI-generated images on Facebook, Instagram and Threads, beginning in May.
But for now — and as always — keeping your guard up is key. If you need some more pointers, check out these expert tips on how to spot AI-generated images and avoid getting tricked online.
