Instagram AI content labels are again producing contradictory results, according to user reports collected by The Verge: photographs that users say were not made with generative AI have received an “AI Content” label, while some actual AI imagery has appeared without one. The reports, published September 4, do not establish how often this is happening or identify a technical cause. They do show why a badge meant to explain an image’s provenance should not be mistaken for a final ruling on how it was made.
The Verge reported that some users connected the unexpected labels to work done in Canva, including its Background Remover, or to small blemish-removal edits. Those are reported associations, not a confirmed explanation from Meta or Canva. Other reported cases had no clear trigger.
Meta has not publicly explained the current reports in the material available. The company previously conceded that its earlier labeling scheme could clash with what users reasonably understood a label to mean.
Why is Instagram labeling regular photos as AI content?
Meta’s system is designed to use signals attached to content and disclosures by uploaders, rather than making a definitive visual judgment about every image. In 2024, Meta said it would apply labels across Facebook, Instagram and Threads when it detected industry-standard AI indicators or when a person disclosed that they were uploading AI-generated material.
For images, Meta specifically identified C2PA “AI generated” information and IPTC technical-standard information as signals its tools were being built to detect. Meta said the goal was to label images from companies including Google, OpenAI, Microsoft, Adobe, Midjourney and Shutterstock as those companies added the relevant metadata to their outputs.
That setup can create an awkward middle ground. In a July 2024 update, Meta said labels triggered by industry indicators did not always meet people’s expectations or provide enough context. It cited minor AI modifications, including retouching tools, that had resulted in a “Made with AI” label. Meta changed that wording to “AI info.”
By September 2024, the company said it would put the AI-info notice inside a post’s menu when it detected that content had only been modified or edited with AI tools. It said labels would remain displayed for content detected as generated by an AI tool, and would identify whether the basis was an industry-shared signal or a user disclosure.
Can Instagram identify every AI-generated image?
No. Meta said in its 2024 announcement that it could not identify all AI-generated material and that invisible markers can be removed. It also said it was developing classifiers intended to spot AI-made content that lacks those markers, but that was a stated development effort, not a claim that the problem had been solved.
The current reports therefore point to a two-sided limitation: a label may reflect a detected indicator or an uploader’s disclosure, while an absent label does not establish that an image was made without AI. Meta has advised people assessing content to also consider whether the sharing account is trustworthy and whether details in the image or audio seem unnatural. That is less tidy than a universal detector, because no such detector is what Meta described building.
For creators and viewers, the practical implication of the reported false positives and false negatives is modest but important: treat Instagram’s label as one piece of context, not conclusive proof of either synthetic origin or human authorship.
This story draws on original reporting from The Verge.
