Over the past few weeks, people scrolling Instagram have started noticing a small “Made with AI” badge in the top-left corner of reels and posts — on images that plainly weren’t generated by an AI tool. A woman’s portrait. A sports celebration photo. A haircut result shot in a studio. Real photography, tagged as if it came out of a text prompt.
As TechRadar reported, creators including Peter McKinnon pushed back publicly, and podcasters Colin and Samir put the question directly to Instagram’s head, Adam Mosseri.
What Mosseri actually said
His answer pointed at the file, not the pixels:
“It’s just the metadata on the image. People are doing benign things in Photoshop that are technically using AI. And then we’re just respecting the file’s mark saying ‘Hey, AI was used to make this.’ We have to figure this out because clearly this is frustrating and not the ideal experience.”
In plain terms: Instagram is reading Content Credentials (C2PA) information that Adobe writes into a file’s metadata whenever a generative feature touches it — even something as small as one Generative Fill cleanup on a blemish, or a content-aware crop. That flag then rides along in the exported JPEG, and Meta’s systems surface it as a binary “made with AI” badge with no distinction between “fully AI-generated” and “one small AI-assisted touch-up on a real photo.”
Former White House photographer Pete Souza reported the same thing happening to his own work, and TechRadar noted a photo of the Kolkata Knight Riders’ IPL win getting the same tag — both ordinary camera photos that had simply passed through an editing tool with generative features somewhere in the pipeline.
What Reddit and the Adobe community found
The reaction spread beyond Instagram’s comment section. A long Adobe Community thread filled up with working photographers who felt blindsided — many said they had no idea a single Generative Fill cleanup would follow the file all the way to a public “made with AI” label, and that it undercut how clients perceived real, paid photography work.
One workaround that circulated in that thread: opening the exported JPEG in an unrelated program like Microsoft Paint and re-saving it, which strips the embedded Content Credentials data because Paint never reads or rewrites it. It’s a blunt tool for the job, but it points at the same underlying fact Mosseri described — the badge is driven by information sitting in the file, not by anything Instagram can see in the pixels themselves.
On Instagram itself, replies to Colin and Samir’s post argued for a vocabulary fix rather than scrapping the label. User @voust put it this way:
“There should be ‘edited with AI and made with AI’... ‘made’ kind of implies the whole image is artificial.”
Meta has moved a little in that direction before: in mid-2024, after an earlier round of the same complaint, it renamed the badge from “Made with AI” to the softer “AI info,” per TechCrunch. The wording softened; the underlying metadata-reading mechanism — and the same false positives on lightly edited real photos — did not.
Why a binary tag is a hard problem
TechRadar’s piece also raises the flip side: Instagram can’t simply drop metadata-based detection, because plenty of fully AI-generated images get uploaded with no metadata at all — a flat screenshot, for instance — and sail through untagged. Mosseri acknowledged this directly: “There will be content on all platforms that is made with AI and is not marked with AI because the person who made it specifically tried to avoid that.” Metadata is an imperfect signal in both directions — it over-flags careful, real photography and under-flags anything stripped of its file information on purpose.
What’s driving the false positives
Content Credentials / C2PA metadata that Photoshop and similar tools write into a file whenever a generative feature touches it, however lightly.
What a rename didn’t fix
“Made with AI” became “AI info,” but the detection logic reading embedded metadata is unchanged.
If you edited a real photo and don’t want that metadata along for the ride
This is exactly the file-level problem AI Label Removeris built for. If a photo is genuinely yours — shot on a camera or phone, lightly retouched — and you don’t want the Content Credentials, EXIF, GPS or other embedded information traveling into the copy you upload, you can scan the file, see in plain language what’s embedded in it, and export a clean copy with the supported information removed — entirely in your browser, with nothing uploaded to a server.
That said, it’s worth being precise about what this does and doesn’t solve. Removing embedded metadata addresses the exact signal Mosseri described. It cannot guarantee Instagram won’t apply a label anyway — platforms can still use disclosures, classifiers or other context a file cleaner has no visibility into. And it’s not a way to misrepresent AI-generated content as real: if a platform asks you to disclose AI use, answer honestly regardless of what a file scan shows.
Scan a photo now