Instagram’s AI detection system is misfiring again, and this time the false positives are hitting ordinary photos that never touched a generative AI tool. Users across Threads and Instagram report that Meta has been automatically stamping an “AI Content” label on images edited with Canva’s Background Remover — sometimes for removing what one user described as “a speckle” — and even on photos taken on an iPhone and lightly edited in Apple’s Photos app. The label appeared on recent posts from About Face, the cosmetics brand founded by Halsey, prompting its social manager to insist: “This photo was taken on my iPhone and then slightly edited by myself in the photo app. Our brand uses real artist[s] and real people to create everything you see across all our platforms.”
This isn’t the first time Meta’s labeling has gone sideways. A similar wave of erroneous “Made by AI” tags hit photographers in 2024, just months after the feature launched. Back then, the system was catching images whose Adobe metadata indicated they’d been retouched with generative tools, even when the edits were trivial. Meta promised to refine its approach to better reflect “the amount of AI used in an image,” but the company has stayed vague about exactly what metadata it scans and when. Meta didn’t respond to a request for clarification on the current problems.
At least some of the blame appears to fall on Canva. Content strategist Jess Bruno says Canva told her that some of its assistive AI tools “were being tagged as generative” and that the issue has since been fixed. Canva’s help page now states that Background Remover “doesn’t add Canva’s AI-generated content metadata to your design.” Yet multiple Threads users say their images still get flagged after using the tool, and others report that images edited before the fix never got tagged at all — raising questions about whether C2PA metadata is even being applied consistently.
The deeper problem is that Instagram’s labels are now doing the opposite of what they’re for. Assistive tools like background removal have used machine learning for over a decade — this is the same category of AI as Photoshop’s object selection, not the text-to-image generation people associate with deepfakes. When legitimate photos get flagged while actual AI imagery slips through, the label stops being a trust signal and becomes noise. If users can’t tell whether “AI Content” means something meaningful, they’ll eventually stop paying attention to it altogether — which is arguably worse than having no label at all.
