Instagram's AI Detection System Mislabels Photos While Missing Actual AI Content

Meta introduced "AI Content" labels on Instagram to help users identify synthetically generated images at a glance. Instead, the system has become unreliable — automatically marking legitimate photos while overlooking actual AI-generated content, turning the whole labeling scheme into a source of frustration rather than trust.

The Labeling Chaos#

Users report that the system incorrectly flags ordinary photos for using basic editing features. Images edited with Canva's Background Remover tool trigger the label even when no generative AI was involved. Similarly, minor touch-ups in the iPhone Photo app—a few blemish fixes or small adjustments—have resulted in entire images being marked as AI-generated. Meanwhile, genuinely AI-created or heavily manipulated images frequently slip past detection entirely.

One cosmetics brand founded by the singer Halsey had multiple product photos flagged despite the social manager clarifying that the images were shot on iPhone and slightly edited in standard photo editing apps. The inconsistency has left creators and brands bewildered, unsure whether their legitimate edits will be flagged.

A Pattern of Failure#

This isn't Meta's first stumble with AI detection. A similar crisis unfolded in 2024 when Instagram over-aggressively labeled photos based on metadata from Adobe tools, even when only minimal AI-assisted retouching was applied. Meta promised to improve the system and adjust labeling to reflect "the amount of AI used in an image," yet recent confusion suggests those fixes haven't resolved the core issues.

The root cause remains murky. Meta hasn't publicly disclosed which specific signals its detection system scans for, though the company previously indicated it checks for IPTC and C2PA metadata—technical markers that can indicate AI involvement.

Where Canva Comes In#

Some of the problem traces back to Canva's tools. The design platform told one content strategist that its assistive AI features were incorrectly being tagged as generative by Instagram's system. Canva stated it has corrected the tagging on its end, though users continue reporting that images edited with Canva's Background Remover still get flagged even after the claimed fix.

Notably, the Background Remover uses machine learning for object selection—assistive AI technology that's been standard in Photoshop and other tools for over a decade. It's fundamentally different from generative text-to-image models often associated with deceptive synthetic content.