AI Themed Met Gala? Rihanna and Katie Perry Shrug It Off But Deepfakes Are No Laughing Matter

Key Takeaways

• Neither Katy Perry nor Rihanna attended this year’s annual Met Gala.
• However, AI-generated images of the 2 singers appeared to show them in attendance.
• The deepfake images were seen by millions of people online, highlighting how easily they can spread, often without being labeled as fake.

Although Katy Perry and Rihanna have both attended the Met Gala in the past, neither was at this year’s event on Monday, May 6. But you wouldn’t know that from photos that circulated online. 

Deepfake images purporting to show the singers on the red carpet were so believable that even Perry’s mother fell for them. 

Met Gala Gets an AI Makeover

In an Instagram post attending to various deepfake images that appeared to show her at the Met Gala, Perry shared a screenshot of a message from her mother commenting on her “gorgeous gown.”

However, the singer wasn’t actually in attendance. “Mom the AI got you too, BEWARE!” she responded.

Other celebrities who received the Met Gala deepfake treatment include Rihanna and Lady Gaga.

The former, who is a regular at the annual charity event, reportedly had to cancel due to illness. Nevertheless, an AI-generated image of her in a structural cream dress adorned with birds and flowers racked up millions of views on social media.

Meanwhile, although less convincing, a deepfake image depicting Lady Gaga looking a bit like a wedding cake was also circulated online. 

Social Media Platforms Respond to Rising Deepfakes

Although the recent Met Gala deepfakes are relatively innocuous, more malicious applications of the technology have been well-documented.

From fraud to election interference, modern generative AI tools have made it easier than ever to fool people with digital images. 

Although some of the deepfake photos triggered social media platforms’ AI labeling systems, not all of them did, suggesting that monitoring the emerging technology is still a challenge.

Digital Watermarking a Potential Solution

Current approaches to deepfake labeling often rely on community- and self-reporting. However, there is also a growing emphasis on digital watermarking.

For example, OpenAI embeds digital watermarks in DALL-E outputs’ metadata. But such safeguards can be easily bypassed. more sophisticated solutions can hide identifying markers in the deepfake images themselves. 

With synthetic media more prolific than ever, such technologies “can solve mistrust at scale,” Eric Wengrowski, the CEO of digital watermarking startup Steg.ai told CCN.

In the end, trust is exactly what convincing deepfakes threaten and why technologies that increase transparency are needed. Perry’s warning to her mother rightly suggests that the online space can be dangerous when AI-generated images can fool even family members.