Glonce

Instagram's AI Content label keeps tagging real photos, missing real AI ones

productsThe Glonce editorsThe Verge🔥56Glonce rating56Impact50Novelty35Relevance80Surprise40Players70Credibility85
Instagram's AI Content label keeps tagging real photos, missing real AI ones

Instagram's visible "AI Content" label is meant to let people spot synthetic images at a glance, but for several weeks users have reported it firing on photos they never touched with generative AI. Many blamed simple editing tools: several people said the tag appeared after using Canva's Background Remover, including one user who lost an entire photo to the label after removing just "a speckle." A similar mislabeling wave hit Instagram in 2024 under a "Made by AI" label, a few months after Meta said, that February, it would start scanning images for IPTC and C2PA metadata; Meta later promised to better reflect "the amount of AI used in an image" but has stayed vague about its methods and did not respond to a request for clarification this time either. Content strategist Jess Bruno said Canva told her some of its assistive AI tools "were being tagged as generative" and that its tools now tag correctly; a note on Canva's help page says the Background Remover "doesn't add Canva's AI-generated content metadata to your design." Yet some Threads users say the tool still triggers the label even after Canva's claimed fix, while others say images edited before the fix were never tagged and never carried C2PA metadata at all. Background removal tools use the same assistive machine learning that Photoshop has used for over a decade, not the generative text-to-image systems the label is supposed to catch. The pattern goes beyond Canva. One user found the tag applied to an image that had been "poisoned" (subtly altered to make it useless for AI training) while an unpoisoned version of the same content went untagged. Recent Instagram photos from About Face, a cosmetics company founded by singer Halsey, were also automatically labeled; the brand's unnamed social manager replied that the photo "was taken on my iPhone and then slightly edited by myself in the photo app," adding the company uses real artists and real people throughout its content. Only a handful of iPhone Photos tools should add generative-AI metadata: Spatial Reframing, Extend, and the updated Clean Up added in iOS 27, all of which use Apple Intelligence and embed Google's SynthID watermark. But the flagged images did not carry SynthID, according to Google's Gemini-based verification, leaving no clear explanation for why they were tagged at all. The Verge's own testing found the opposite failure. Running photos edited or generated with Canva's Background Remover, Photoshop's background eraser, Adobe Firefly, Google's Nano Banana model and Apple Intelligence features, including some confirmed to carry C2PA and SynthID signals, none were labeled by Instagram. The only trigger that reliably worked was Meta's own generative AI app. The author then posted the images to a brand-new Instagram account with minimal profile information, publishing them in quick succession to mimic bot-like behavior, precisely the kind of account Meta's recently announced crackdown on AI accounts was said to target. The account was never flagged as AI-generated, and the pictures went unlabeled for almost two weeks. Meta did not explain what signals the AI Content label is actually built to detect.

Key facts

  • Instagram's "AI Content" label has been firing on ordinary photos for several weeks, echoing a 2024 mislabeling wave under a "Made by AI" tag.
  • Multiple users traced false tags to Canva's Background Remover; Canva says its tools now tag correctly, but some users report the problem persists.
  • A cosmetics company founded by singer Halsey, About Face, had recent photos tagged despite saying they were shot on an iPhone and only lightly edited by a person.
  • Flagged images did not carry Google's SynthID watermark even when Apple Intelligence tools that embed it were suspected as the cause.
  • The Verge's own tests found the label applied only to images made with Meta's own AI app, and a bot-like test account mimicking an AI farm went completely untagged for nearly two weeks.

Why it matters

Instagram's AI label exists to let people tell at a glance whether an image is synthetic, but the reporting shows it doing close to the opposite: flagging real photos edited with ordinary tools while letting genuine AI-made or AI-poisoned images through untagged. Meta has stayed vague about what signals the system checks for since introducing it, and did not respond to a request for clarification here, so the label's actual detection logic remains unknown even after a second public mislabeling episode.

Who it affects

Instagram users who edit photos with common tools like Canva's Background Remover or iPhone Photos features risk having ordinary pictures publicly marked as AI-made. Brands and creators, such as About Face, face the same reputational problem when a mislabel implies AI use they say did not happen. Canva and Apple are drawn in as well, since their assistive tools are being blamed for triggering a label meant for a different category of software.

How to use it

There is no reliable way for a user to avoid the mislabeling, since the article found no consistent cause: Background Remover triggers it for some people and not others, and pre- and post-fix edits from Canva produced inconsistent results. The only tool the author found that consistently triggered the label was Meta's own generative AI app; every other tool tested, including ones confirmed to embed C2PA or SynthID signals, went unlabeled.

How solid is it

The account rests on named on-the-record testimony (Jess Bruno, About Face's social manager) plus the author's own controlled tests across five separate tools and a purpose-built Instagram account. Meta, Canva and About Face were all approached for comment; none responded to The Verge directly, though Canva's explanation reached the reporter through content strategist Jess Bruno and a help page. The consistent, reproducible failure in the author's own testing across multiple platforms strengthens the account, even though Meta's silence means the underlying cause is not confirmed.

Risks and caveats

No figures exist for how many users or images are affected, and no start date is given beyond "over the last few weeks." Meta never confirmed a cause, so explanations offered by Canva and inferred from Apple Intelligence tools remain partial and unverified by the company running the system. It is also unclear whether Canva's claimed fix is complete, since some Threads users report the mislabeling continuing afterward.

“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.”

— About Face's unnamed social manager