We recently asked Meta if it trains its AI on the photos and videos users take with Ray-Ban Meta smart glasses. Initially, the company did not have much to say.
Since then, Meta has added a little more color to TechCrunch.
This means that any images you share with Meta AI can be used to train that AI.
“[I]”Where Multimodal AI is available (currently in the United States and Canada), images and videos shared with Meta AI may be used to make improvements in accordance with our Privacy Policy,” said Emil, Meta Policy Communications Manager. Vazquez said in an email to TechCrunch.
In a previously emailed statement, a spokesperson said photos and videos taken with Ray-Ban Meta will not be used by Meta for training unless the user sends them to the AI. I made it clear. However, when you ask Meta AI to analyze those photos, they fall into a completely different set of policies.
In other words, the company is using its first consumer AI device to create a massive stockpile of data that can be used to create an even more powerful generation of AI models. The only way to “opt out” is to not use Meta's multimodal AI features in the first place.
Ray-Ban Meta users may not realize that they are providing Meta with a large number of images (perhaps the inside of their home, loved ones, personal files, etc.) in order to train a new AI model. This effect is a concern. According to a Meta spokesperson, this is obvious in Ray-Ban Meta's user interface, but company executives either didn't initially know or didn't want to share these details with TechCrunch. We already knew that Meta was training its Llama AI model on everything Americans publish on Instagram and Facebook. But now, Meta has expanded this definition of “public data” to include anything people view through smart glasses or ask AI chatbots to analyze.
This is especially relevant right now. On Wednesday, Meta began rolling out new AI features that make it easier for Ray-Ban Meta users to invoke Meta AI in a more natural way. This means that users are more likely to send new data to Meta AI that can also be used for training. Additionally, the company announced new live video analytics capabilities for Ray-Ban Meta at its 2024 Connect conference last week. This essentially sends a continuous stream of images to Meta's multimodal AI model. Mehta said in a promotional video that the feature allows you to look through your closet, use AI to analyze it all, and choose your outfit.
What the company doesn't advertise is that it also sends these images to Meta for model training.
A Meta spokesperson pointed to TechCrunch's privacy policy, which clearly states that “your interactions with AI features may be used to train AI models.” This appears to include images shared with Meta AI through Ray-Bans smart glasses, which Meta has not yet revealed.
The spokesperson also pointed TechCrunch to Meta AI's terms of service, which state that by sharing images with Meta AI, “Meta uses AI to analyze images containing facial features.” This means that you agree to do so.”
Meta just paid $1.4 billion to the state of Texas to settle a lawsuit related to the company's use of facial recognition software. The lawsuit revolved around a feature Facebook introduced in 2011 called “tag suggestions.” By 2021, Facebook made this feature an explicit opt-in option and deleted the biometric information of billions of people it had collected. Notably, some of Meta AI's imaging capabilities have not been released in Texas.
Elsewhere in Meta's privacy policy, the company says that by default it will also save all transcriptions of audio conversations with Ray-Ban Meta to train future AI models. As for the actual audio recordings, there are ways to opt out. When logging into the Ray-Ban Meta app for the first time, users can choose whether they can use audio recordings to train Meta's AI models.
It's clear that Meta, Snap, and other tech companies are pushing smart glasses as a new computing form factor. All of these devices have cameras that people wear on their faces, and most of them are powered by AI. This revisits a plethora of privacy concerns first heard in the days of Google Glass. 404 Media reported that some college students have already hacked Ray-Ban meta glasses to reveal the names, addresses, and phone numbers of people looking at them.