Instagram users are reporting that Meta’s “AI Content” labels are being attached to ordinary photographs, while some fully AI-generated images appear to be escaping detection.
The tagging system is intended to help people identify synthetic material, but complaints in recent weeks suggest it may be confusing basic editing tools with generative artificial intelligence. Users have described labels appearing after removing a background in Canva or making minor blemish corrections.
One Threads user said Instagram applied the label “every time there is bg remover involved”. Another said a Canva edit to remove a single speck from a photograph resulted in the entire image being marked as AI-generated.
The problem appears to involve more than one tool. Some users say images have been labelled despite no known use of Canva’s Background Remover, while others have found that one version of a picture was marked and an almost identical version was not.
Canva has acknowledged that some of its assistive tools were incorrectly being identified as generative AI. The company told content strategist Jess Bruno that the tools were now tagging content correctly, although users have continued to report problems with images edited using Background Remover.
A message on Canva’s help page says the feature does not add the platform’s AI-generated content metadata to a design. It is not clear when that wording was added, and Canva has not provided further comment.
Background removal and similar functions can use machine learning, but they are generally regarded as assistive editing tools rather than the text-to-image systems associated with fabricated or deceptive imagery.
Instagram’s unclear AI detection system
Meta introduced its “Made by AI” labelling system in 2024 after saying it would examine metadata, including IPTC and C2PA information, to identify whether generative tools had been used to create or alter an image.
The company has previously admitted that its system could label photographs after minimal generative editing. It said it would refine the process so that labels better reflected “the amount of AI used in an image”, while also referring to industry-standard signals embedded by technology companies.
Meta has not publicly explained in detail which signals it currently scans for, when those checks take place or how it distinguishes between a photograph that has been lightly edited and one that has been generated from scratch.
Tests described by technology journalist Jess Weatherbed found that images created or edited using Canva, Adobe tools, Google’s Gemini and Apple Intelligence were not consistently labelled by Instagram. Images produced using Meta’s own AI application were the only examples that reliably triggered the label.
In one test, fully AI-generated images made in Google Gemini contained C2PA and Google’s SynthID markers but were not labelled on Instagram. By contrast, images created in Meta AI were identified.
The inconsistency has also affected businesses. About Face, the cosmetics company founded by singer Halsey, said in a reply to a commenter that photographs carrying an AI label had been taken on an iPhone and only lightly edited in the phone’s Photos app.
Some Apple Intelligence features can add generated elements or remove objects from photographs, and may embed provenance information. However, the images identified in the About Face case reportedly did not contain Google’s SynthID watermark, and it remains unclear what prompted Instagram’s warning.
Meta has also said it is taking action against accounts that publish large volumes of AI-generated material. But users’ experiences suggest that the platform’s labels are currently neither a reliable guarantee that an image is synthetic nor proof that it has been created entirely by artificial intelligence.
