Meta Rolls Out AI Feature That Scans Phone Camera Rolls

Meta has introduced a controversial new feature for Facebook users in the United States and Canada that allows artificial intelligence to scan through unpublished photos stored on their phones. The opt-in feature promises to help users discover and share memorable moments by automatically identifying noteworthy images buried among screenshots and everyday snapshots.

The feature works by uploading photos from user camera rolls to Meta’s cloud servers, where AI algorithms analyze the content to suggest creative edits, collages, and video compilations. Meta positions this as a solution for people who capture numerous photos but rarely share them due to time constraints or concerns about whether their content is interesting enough for social media.

Users who enable the feature will see AI-generated suggestions appear in their Facebook Stories and Feed, with additional recommendations accessible through the Memories bookmark section. The company emphasizes that all suggestions remain private until users actively choose to share them, and participants can disable the feature at any time through Facebook camera roll settings.

The privacy implications of this feature have drawn significant attention from technology observers and privacy advocates. While Meta states that media from camera rolls will not be used to improve its AI systems, this protection comes with notable exceptions. The company has confirmed that once users choose to edit suggested content with AI tools or publish the AI-generated creations to Facebook, that media may then be used for AI training purposes.

This data usage policy represents a shift from earlier assurances. When Meta initially tested similar functionality in June, the company declined to commit to never using unpublished private photos for AI training. The current implementation appears to fulfill those earlier reservations about future data use.

Meta’s clarification indicates that simply opting into the feature and allowing photo uploads does not immediately make those images available for AI training. However, the moment users interact with AI editing tools or share the resulting content, their previously private photos enter Meta’s AI development pipeline. The company has indicated it may retain some uploaded data for periods exceeding 30 days, though it claims this information will not be used for advertising targeting.

This development follows Meta’s acknowledgment last year that it had already trained its AI models on all public photos and text posted to Facebook and Instagram by adult users dating back to 2007. The new feature extends AI access to previously private content, contingent on user actions.

Meta plans to expand testing of this feature to additional countries in coming months.