Facebook tests AI photo suggestions using your private camera roll

Facebook is quietly testing a new feature that taps into users’ personal camera rolls, including photos that haven’t been uploaded to the platform.

The company is prompting users to opt into “cloud processing” when they attempt to create a new Story on the app. If they accept, Facebook gains permission to automatically upload their photos to Meta’s cloud servers to offer AI-powered suggestions like collages, recaps, restylings, or themed visuals.

As per the on-screen message, by tapping “Allow”, users grant Facebook access to analyze their media based on time, location, and subject matter. The company emphasizes that “only you can see the suggestions” and claims that the media will not be used for ad targeting.

What Happens When You Say “Allow”

Saying yes to this feature also means agreeing to Meta’s AI Terms of Service. According to those terms, Meta has the right to analyze your images and facial features using AI, generate new content, and retain and use personal information to personalize AI interactions. This may include summarizing image contents or modifying them, all based on Meta’s discretion.

Interestingly, the AI terms leave some room for concern. They mention that Meta can review user interactions, including conversations, and that such reviews “may be conducted by humans”. The terms, however, do not clearly define what qualifies as “personal information,” except for vague categories like “Prompts, Feedback, or other Content.”

This has raised privacy questions, especially since users are allowing access not just to photos they’ve posted, but also to those still sitting privately in their camera roll.

Hidden Settings, Limited Awareness

Despite the potential implications, user backlash has been minimal. A few Facebook users who encountered the feature while posting Stories expressed confusion, with some noticing their photos had been automatically stylized into effects like anime without direct request. Others turned to Reddit and Facebook groups seeking help to disable the tool.

The settings can be found under Preferences > Camera Roll Sharing Suggestions within the Facebook app. Here, two toggles are present:

• One allows Facebook to suggest photos from your camera roll during app usage.
• The other controls “cloud processing,” the very feature enabling Meta AI to create edited suggestions from your private media.

While the feature feels new, older posts dating back to early 2024 suggest that the rollout began earlier, with several users already reporting unexplained AI suggestions from photos in their camera roll.

Privacy Boundaries and AI Access

What’s raising eyebrows is how this new level of access goes beyond Meta’s earlier AI disclosures. Initially, the company stated it would only train AI models on public data, such as posts and comments on Facebook and Instagram. But the camera roll access means private media is now being drawn into the AI ecosystem — even if, for now, Meta claims it’s “not used to improve AI models.”

This is especially relevant given that Meta’s AI terms became enforceable on June 23, 2024, and prior versions of these terms are no longer publicly accessible for comparison, not even through the Internet Archive’s Wayback Machine.

EU users had a deadline of May 27, 2025, to opt out of Meta’s AI training using their public data. But this new feature, involving private photo access, wasn’t part of the original disclosure.

Still in Testing, Limited to the U.S. and Canada

According to Meta spokesperson Maria Cubeta, the tool is still in its testing phase and is currently being trialed in the U.S. and Canada. She emphasized that the suggestions are:

• “Opt-in only”
• “Shown only to the user”
• “Can be turned off at any time”

Cubeta clarified that while media from your camera roll is used to improve content suggestions, it is not used to train AI models, at least within this test phase.

While Meta positions this as a convenience feature aimed at making content sharing easier, it also reflects the increasingly blurry lines between convenience and privacy. With AI embedded deeper into social apps, users must make quick choices, often without fully understanding what they’re consenting to.

Whether this feature stays or disappears depends on how users react. It’s not just about clicks — awareness matters too. As tech giants compete in AI innovation, one question remains: What are we really giving up for those AI-generated photo suggestions?