Instagram’s AI Detection Faces New Challenges
Instagram’s AI labels are designed to help users quickly identify content generated by artificial intelligence. However, recent reports indicate that the system is malfunctioning. Users have noticed that the “AI Content” label is being applied to images that were neither created nor edited with generative AI tools. Meanwhile, genuine AI-generated images are slipping through undetected, leading to widespread distrust in the platform’s labeling system.
The reasons behind these erroneous tags seem to vary. Many users have reported that the label appears on images edited with tools like Canva’s Background Remover or even with minimal touch-ups. One user mentioned that Instagram flagged their photo simply for using a background remover, while another noted that a minor edit resulted in their entire image being labeled as AI-generated. In some instances, the reasons for the tagging are even less clear.
This situation may remind some of a similar issue that arose in 2024 when Instagram introduced a “Made by AI” label. Back then, Meta announced plans to scan images for IPTC and C2PA metadata to determine if generative AI was involved in their creation or modification. However, the company has been vague about its methods. Reports indicated that the detection system flagged images with Adobe metadata suggesting they had been slightly retouched with generative AI tools, even when the changes were minimal. Meta promised to refine its labeling process to better reflect the extent of AI involvement in images, claiming to use “industry standard indicators,” but details on these indicators remain scarce.
The recent labeling errors do not appear to have a direct connection to generative AI usage. While background removal features often utilize machine learning, they rely on assistive AI, similar to tools in Photoshop, rather than the generative models typically associated with misleading AI imagery.
Some of the issues with Instagram’s AI labeling seem to stem from Canva. After raising concerns about the Background Remover, a content strategist reported that Canva acknowledged some of its assistive AI tools were incorrectly tagged as generative. Canva stated that its tools are now tagging correctly, and their help page mentions that using the Background Remover does not add AI-generated content metadata to designs. However, it remains unclear if this information is new, and Canva has not provided further comments.
Despite Canva’s claims, some users on Threads continue to report that images edited with the Background Remover are still being tagged. Others noted that images they edited before the fix were not tagged, nor did they contain the C2PA metadata that Meta supposedly uses to identify AI-generated content.
In my own tests to determine what triggers the AI label on Instagram, I found that it was only applied to images I edited or created using Meta’s AI app.
There are also instances where the AI tag was applied without any use of Canva’s Background Remover. One user discovered that an image altered by a system designed to degrade AI training data was tagged, while an unaltered version of the same image was not.
The AI Content label was also automatically applied to recent posts from a cosmetics brand founded by a well-known singer. When questioned about the AI tag, the brand’s social media manager clarified that no AI was involved in creating the images, stating that they were taken on an iPhone and lightly edited. They emphasized that their brand relies on real artists and people for all content. I reached out for clarification on the specific tools used in the iPhone Photos app, but the brand has not yet responded. Only a few tools in the app, such as Spatial Reframing and Clean Up, are known to add generative AI metadata, and the tagged images reportedly do not contain this metadata.
The confusion surrounding what AI signals Meta is scanning for is palpable. Companies often keep the specifics of their moderation systems under wraps to prevent exploitation, but users are understandably perplexed by the current situation.
The only clear trigger for the AI Content label appears to be content generated by Meta’s own AI systems. I conducted tests with various tools, uploading images edited with Canva’s Background Remover, Photoshop, and other generative AI features. Despite confirming that some of these images included C2PA and SynthID signals, none were labeled by Meta.
I created a new Instagram account with minimal information and posted the images in quick succession, attempting to mimic the behavior of a bot or AI farm. Despite Meta’s recent crackdown on AI accounts, my profile and images remained untagged, even after being public for nearly two weeks.
These results have left me questioning the logic behind how Meta applies “AI Content” tags. Intended to enhance trust in what users see on Instagram, the current state of the labeling system has made me skeptical of the labels themselves.
