Instagram's Meta artificial intelligence (AI) content detection tool has been spotted labelling real images as “AI-created.” One of the mislabelled posts was from the official Instagram account of Kolkata Knight Riders, who recently won the Indian Premier League (IPL) 2024 cricket tournament. One image posted by the account, showing the team holding up the trophy, was labelled as AI-generated by the platform. Several photographers on the social media platform have experienced the same issue.
In February, Meta said it was rolling out an AI-generated content detection feature to help protect users from misinformation and highlight instances of deepfakes (images or videos that are AI-generated or digitally altered to resemble another person, place, or event). The feature recently rolled out to Instagram, where real photos appear to be labeled as AI-generated content as well. Currently, these labels only appear on the iOS and Android apps, not on the web.
A photo of Kolkata Knight Riders labelled as 'created by AI'
Photo credit: Instagram/kkriders
The KKR photo is one of the most famous examples of this mistake, but many other mislabeled images have been called out by users, including former White House photographer Pete Souza, who posted a photo of an old basketball game.
After being incorrectly labeled, he edited the caption, writing, “I don't understand why Instagram is using the label 'Created with AI' on my post. No AI is used on my photos,” and stressed that he couldn't uncheck the label because the platform had added it multiple times.
Frustrated users also flooded threads on Meta, a text-based social media platform, and began calling out the issue, tagging Instagram head Adam Mosseri. “No one at Facebook or Instagram has any idea what would trigger a 'Made with AI' label for any photographer or artist. Mosseri clarifies that they're just reading the C2PA label, but no one knows how to avoid it,” one user wrote.
Earlier, Meta's president of global affairs, Nick Clegg, said the company was “working with industry partners to develop common technical standards that will signal when content has been created using AI,” and claimed the detection tool can correctly label images from Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock.
However, the implementation of the feature appears to be flawed: PetaPixel reports that simply removing small specks in an image using Adobe's AI-powered Generative Fill resulted in the image being tagged as “Made by AI,” but non-AI tools like the Spot Healing Brush tool and the Clone Stamp tool did not add the label, despite producing the same results.
The magazine also discovered that when an image that had previously been labeled as AI-generated was reloaded into Photoshop, copied and pasted into a black document, and saved, the AI label no longer appeared.
Meta spokesperson Kate McLaughlin told The Verge that the company is currently evaluating its approach, taking into account recent user feedback. The Verge quoted McLaughlin as saying, “Because we rely on industry-standard metrics that other companies incorporate into the content of their tools, we are actively working with these companies to improve our processes and ensure that our labeling approach aligns with our intent.”
