Meta is reportedly exploring a new feature that could allow its AI models to access private, unpublished photos from users' camera rolls. This development has sparked significant privacy concerns, as it suggests a potential shift in how the tech giant utilises user data for AI training, moving beyond publicly shared content.
Meta's New AI Initiative Sparks Privacy Debate
Facebook users have recently encountered pop-up messages when attempting to post to Stories, inviting them to opt into "cloud processing." This feature would enable Facebook to regularly upload media from a user's camera roll to its cloud. The stated purpose is to generate creative content suggestions such as collages, recaps, AI restyling, or themed content like birthdays or graduations.
However, opting into this feature means agreeing to Meta AI terms, which grant Meta the right to analyse "media and facial features" from these unpublished photos, including metadata like the date taken and the presence of people or objects. Crucially, users would also grant Meta the right to "retain and use" this personal information.
Shifting Data Utilisation for AI Training
For years, Meta has trained its AI models using billions of public images uploaded to Facebook and Instagram. The company recently acknowledged scraping data from all content published on these platforms since 2007 for generative AI training, stating it only used public posts from adult users over 18. However, the definition of "public" and "adult user" in 2007 has remained vague.
This new "cloud processing" feature represents a potential expansion into previously private data. While Meta claims the current test does not use these photos to train or improve its AI models, the terms of service do not explicitly exempt unpublished photos from being used as training data in the future. This contrasts with Google Photos, which explicitly states it does not train generative AI models with personal data from its service.
Key Takeaways
- Opt-in Feature: The "cloud processing" feature is presented as opt-in, allowing users to choose whether to grant Meta access to their camera roll.
- Data Analysis: If opted in, Meta's AI can analyse media, facial features, dates, and objects within unpublished photos.
- Retention Rights: Users grant Meta the right to retain and use this personal information.
- Privacy Concerns: The lack of clarity regarding future AI training use of these private photos raises significant privacy concerns.
- User Control: Users can disable camera roll cloud processing in their settings, which will also initiate the removal of unpublished photos from the cloud after 30 days.
User Control and Future Implications
While Meta states the feature is "very early" and entirely opt-in, the implications for user privacy are significant. The company's public affairs manager, Ryan Daniels, clarified that the current test does not use these photos for AI model training. However, Meta has not confirmed whether this will remain the case in the future, leaving a grey area in its AI usage terms, which have been in place since 23 June 2024.
Despite assurances that the feature only retrieves 30 days' worth of camera roll data at a time, Meta's terms indicate that suggestions based on themes (e.g., pets, weddings) may include media older than 30 days, suggesting longer data retention for certain purposes. Users do have the option to turn off camera roll cloud processing in their settings, which will also lead to the removal of unpublished photos from the cloud after 30 days.
This development highlights a growing tension between technological advancement in AI and the safeguarding of individual privacy, particularly concerning personal data that was previously considered private and not intended for public consumption.
