Instagram's 'AI Content' label keeps flagging real photos. Here's what trips it.

Open Canva, tap Background Remover, and post the photo. Instagram tags it "AI Content" anyway, and the badge stays on the post. This is the second time in two years that Meta's labeling system has misfired in the same direction, and the company has not said what is different this round.

Instagram is once again tagging everyday photos as "AI Content", even when no generative model touched them. The Verge reports that users on Threads have been seeing the badge on images that only ran through Canva's Background Remover, minor blemish-fixers, or other machine-learning editing tools that are not generative AI. One Threads user said the label appears "every time there is bg remover involved." Another reported that removing "a speckle" with Canva was enough to get the entire image flagged on Instagram.

The 2024 version of this story was about Adobe's retouching tools. Meta's February 2024 announcement said the company would scan images for IPTC and C2PA metadata to determine provenance, and promised the resulting label would reflect "the amount of AI used in an image." The rollout over-flagged Adobe-retouched images and substantively identical photos, drawing the same complaints the platform is hearing again. Meta walked the wording back, said a fix was incoming, and never published a follow-up.

What is different about the 2026 detector is not public. Meta did not respond to The Verge's request for clarification on the current incident, and the company has not described what metadata signals the system reads, what threshold separates "AI Content" from a clean photo, or whether a background remover is supposed to count. A background remover is a deterministic computer-vision pipeline. It does not synthesize pixels the way a diffusion model does. If Meta's system flags it, the label is a definitional choice about what counts as AI in a photograph, and the company has not shared that definition.

The labeling question has moved from product decision to compliance commitment. In July 2026, Meta signed the EU AI Act Code of Practice on Transparency of AI-Generated Content, a voluntary commitment that aligns platform labeling with the bloc's regulatory expectations. The signature puts the 2026 detector under the same scrutiny the 2024 detector escaped.

Instagram's separate policy on AI-generated influencers, which demotes accounts that fail to disclose synthetic personas, sits on top of the same labeling stack. If the on-image badge is unreliable, the disclosure regime above it is unreliable too.

The practical takeaway for anyone posting through the app: read the badge as a hint, not a verdict. A Canva background removal, a Photoshop Generative Fill on a single blemish, or any tool that calls itself "AI-powered" in marketing copy can trip the label. So can images that came out of a real camera and went through nothing more than a color grade. Until Meta publishes what its detector is actually measuring, the "AI Content" tag tells a viewer the photo went through a machine-learning pipeline somewhere. It does not tell them the image is synthetic.

Meta's 2024 promise was that the label would reflect "the amount of AI used in an image." Two years on, the company has not said whether the 2026 version honors that promise, and it has not committed to a date for the next public update.