- Meta launched its new AI sticker feature last week.
- Users have found ways to generate wildly inappropriate images with the tool, and are sharing the results online.
- One academic hit out at Meta’s AI safeguards in a blog post on Sunday.
Meta launched the AI stickers and other features like celebrity AI assistants on September 27.
The company said this new feature allows users to turn “text prompts into multiple unique, high-quality stickers in seconds” on Meta apps like Instagram and Messenger.
But the stickers are already drawing ridicule online after people started trying to figure out ways to generate wildly inappropriate images using the tech.
X user @pioldes shared on Wednesday examples of imagery the new AI tool could create, like a child bearing a gun and a lewd image of Canadian Prime Minister Justin Trudeau.
found out that facebook messenger has ai generated stickers now and I don’t think anyone involved has thought anything through pic.twitter.com/co987cRhyu
— defend trans rights🏳️⚧️ – podesbiens.bsky.social (@Pioldes) October 3, 2023
The post has since racked up over two million views on X, the platform formerly known as Twitter, and soon kicked off a slew of other posts from users sharing their own stickers online.
Though certain words appear to be blocked by the feature, users found ways to bypass these blocks with typos or creative prompts.
The AI-generated stickers are powered by Llama2, Meta’s rival to ChatGPT, and were rolled out to select English-language users. The company wrote in a blog post: “We’re rolling out our new AIs slowly and have built in safeguards.”
However, Tama Leaver, a professor of internet studies at Curtin University, hit out at Meta’s AI safeguards in a separate blog post on Sunday.
Meta claims that their new ‘AI experiences’ “have built in safeguards” so I tested out just how safe the new Instagram ‘AI Stickers’ really are, and how I found showed these tools really aren’t ‘safe’ at all: https://t.co/LZONhtX5Bj
— Tama Leaver ➡️ @email@example.com (@tamaleaver) October 1, 2023
“If part of Meta’s safeguards are blocking the term ‘naked’, but their AI is producing naked figures all the same, there are clearly lingering questions about just how effective these safeguards really are,” he wrote.
Problematic AI-generated content isn’t unique to Meta, however.
In December, the popular AI avatar generator app Lensa faced claims of sexualizing and racializing user avatars. Users said the app sometimes generates nudity in its images despite rules against adult content.
Representatives for Meta, Disney, Nintendo, Sega, and Trudeau did not immediately respond to requests for comment from Insider, sent outside regular business hours.