> ## Content Index
> Fetch the complete content index at: https://www.metatalks.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Instagram tags ordinary photos as AI-made while letting actual AI images through
- URL: https://www.metatalks.ai/instagram-tags-ordinary-photos-as-ai-made/
- Published: 2026-09-09T10:23:00.000Z
- Updated: 2026-09-09T10:22:59.000Z
- Author: Al
- Tags: News, AI Detection, #newswire

**AI-generated pictures carrying standard watermarks posted with no label at all in The Verge's testing. The one thing that reliably set the tag off was Meta's own AI app.**

Instagram's automatic AI Content label is now wrong in both directions: users report the tag landing on pictures they neither created nor altered with generative AI, while genuine AI imagery passes through with no label at all.

[The Verge](https://www.theverge.com/ai-artificial-intelligence/989617/instagram-ai-content-label-confusion?ref=metatalks.ai) tested pictures altered or created with tools from Canva, Photoshop, Adobe Firefly, Google's Gemini and Apple Intelligence. Some were made entirely by AI and verified as carrying C2PA and SynthID signals. Meta labeled none of them. Across the tests one kind of image did draw the tag: pictures edited or generated in Meta's own AI app.

The uploads went out from a brand-new account posting in quick succession, imitating a bot or an AI content farm as closely as it could. Two weeks on, and after Meta announced a crackdown on AI-generated accounts, neither the profile nor the pictures carried a tag.

Errors run the other way as well: users say the label has appeared on pictures that were only put through background removal or given minor blemish fixes.

After the content strategist Jess Bruno raised it with Canva, the company told her it had been marking some of its assistive AI tools as generative when they were not, and that it had fixed this. People on Threads say such images are still being labeled, and Canva has not commented.

Background removal does often lean on machine learning, but that is the assistive variety Photoshop has used to select and erase objects for over a decade — not the text-to-image system people have in mind when they worry about AI pictures.

The tag has also turned up on pictures where Canva's tool was reportedly [never used at all](https://www.instagram.com/p/DbECl%5FKCBlK/?ref=metatalks.ai). In one case Meta labeled a photo that had been poisoned — subtly altered so that models trained on it come out worse — and left an unpoisoned copy alone. Photos from the cosmetics brand About Face were tagged too; its social manager said they were shot on a phone with no AI involved, and a check through Google's verification found no SynthID watermark.

Instagram hit the same problem in 2024, months after the label was introduced. Meta said then that it would change the tag to better match how much AI had gone into a picture, and that it relied on signals other companies embed in what their tools produce.

Which signals it scans for now, and when, it has not said. It did not respond to The Verge's request for clarification.