Instagram's AI Labels Go Haywire

Instagram's visible AI content labels, intended to identify synthetically generated material, have recently begun misfiring. Users are reporting that the system is inaccurately applying an "AI Content" label to images not created or edited with generative AI tools.

This issue appears to stem from various causes, including the use of editing tools like Canva's Background Remover. Even minor edits, such as removing a speck from a photo, have resulted in the entire image being flagged.

Previously, Meta announced in February 2024 that it would scan images for metadata proving AI manipulation. However, the system has a history of erroneous tagging, sweeping up pictures with even negligible AI retouching.

Features like background removal, while utilizing machine learning, are distinct from the generative AI tools typically associated with deceptive content. Meta's detection methods remain vague, and the company did not respond to requests for clarification.

Recent tests indicate that only images edited or fully generated using Meta's own AI app consistently trigger the "AI Content" label. Other tools, including those from Adobe, Google, and Apple, did not result in labeling, even when they embedded AI metadata.

The erroneous labeling, alongside the apparent bypassing of genuine AI content, has left users questioning the reliability of Instagram's AI detection system.