Meta Remains Silent on Whether AI is Trained Using Photos from Smart Glasses

Update: Meta has shared further insights regarding the training of its AI on images captured by Ray-Ban Meta glasses. For more details, click here.

Meta’s AI-enhanced Ray-Ban glasses are equipped with a subtle front-facing camera that not only captures images upon request but also automatically takes photos in response to specific keywords like “look.” As a result, these smart glasses store a significant number of images, both intentional and incidental. However, the company has not committed to ensuring these images will remain private.

When we inquired if Meta intends to use images from Ray-Ban Meta users to train its AI models, as it does with public social media content, the response was ambiguous. “We’re not publicly discussing that,” Anuj Kumar, a senior director overseeing AI wearables at Meta, stated during a video interview.

Mimi Huggins, a Meta spokesperson on the call, added, “That’s not something we typically share externally.” When pressed for clarification on whether these images would be included in training data, Huggins replied, “we’re not saying either way.”

This uncertainty raises concerns, particularly in light of the new AI feature in Ray-Ban Meta that captures numerous passive images. Recently, it was reported that Meta is preparing to introduce a real-time video capability for Ray-Ban Meta, which activates when users say certain keywords. This feature will effectively stream a series of images (essentially live video) into a multimodal AI model, enabling it to provide answers about your surroundings almost instantly.

This process leads to the collection of a large volume of images that the Ray-Ban Meta user may not even consciously realize are being captured. For instance, if you ask the glasses to scan your closet for outfit suggestions, they will take numerous photos of your room and its contents, uploading everything to an AI model in the cloud.

The fate of these images remains unclear, as Meta has not disclosed this information.

Wearing Ray-Ban Meta glasses effectively means you have a camera on your face, which can make others uncomfortable—a sentiment we witnessed with Google Glass. Given this, one might expect the company to assure users, “All your images and videos captured by these cameras will be completely private and stored securely.”

However, Meta has not made such assurances.

Meta has already confirmed it trains its AI models on publicly available data from all American Instagram and Facebook posts. Their interpretation of what constitutes “publicly available data” is quite broad, and it appears we may need to accept this reality. In contrast, other AI providers have more straightforward policies regarding user data. For example, Anthropic has stated that it never trains on a customer's inputs or outputs from its AI models. Similarly, OpenAI maintains it does not train on user inputs or outputs through its API.