We recently asked Meta whether it trains AI on photos and videos that users take on the Ray-Ban Meta smart glasses. The company did not have much to say on the matter at first.

Since then, Meta has offered TechCrunch some color.

In short, any image you share with Meta AI can be used to train its AI.

[I]n places where multimodal AI is available (US and Canada currently), photos and videos posted with Meta AI might be used to fine-tune that AI for our Privacy Policy," wrote Meta policy communications manager Emil Vazquez, in an email to TechCrunch.

In an earlier email statement, a spokesperson clarified photos and videos taken on Ray-Ban Meta aren't used to train Meta unless the user submits them to AI. "Once you ask Meta AI to study them though, though, those photos fall into an entirely different category of policies.".

That is to say, the company is leveraging its first consumer AI device in order to build enormous storehouses of data which might be used to create ever-stronger iterations of the AI models. The only way to "opt out" is to simply not use Meta's multimodal AI features at all.

A worrisome implication of this: Ray-Ban Meta users aren't going to realize that they're uploading tons of images to Meta-including perhaps the interior of their homes, loved ones, or personal files-to train the company's new AI models. Meta's spokespeople say that the interface clearly spells this out on the Ray-Ban Meta user interface; the company's executives simply didn't know this or didn't want to share it with TechCrunch. We already knew Meta trains its Llama AI models on everything Americans post publicly on Instagram and Facebook. Now, however, Meta has broadened the definition of "publicly available data" to include anything people look at through its smart glasses and ask its AI chatbot to analyze.

That's particularly relevant now. On Wednesday, Meta started rolling out new AI features that make it easier for Ray-Ban Meta users to invoke Meta AI in a more natural way-that is to say, users will be much more likely to send it new data that can also be used for training. Another new feature the company rolled out for Ray-Ban Meta at the 2024 Connect conference last week is the live video analysis, which-essentially-feeds a continuous stream of images into Meta's multimodal AI models. In a promo video, Meta said you can look around your closet and analyze the whole thing with AI to pick an outfit.

What the company isn't touting is that you're sending those images to Meta for model training, as well.

Meta representatives pointed TechCrunch to the company's privacy policy, which clearly states, "your interactions with AI features can be used to train AI models." That would seem to include images shared with Meta AI through the Ray-Bans smart glasses, but Meta wouldn't elaborate.

The spokespeople further referred TechCrunch to Meta AI's terms of service, where it is already stated that by uploading images in Meta AI, "you agree that Meta will analyze those images, including facial features, using AI.".

Meta recently settled a court case related to the company's use of facial recognition software by paying the state of Texas $1.4 billion. That case was over a Facebook feature rolled out in 2011 called "Tag Suggestions." By 2021, Facebook made the feature explicitly opt-in, and deleted billions of people's biometric information it had collected. Notably, several of Meta AI's image features are not being released in Texas.

Elsewhere in Meta's privacy policies, it states the firm also retains all transcriptions of all your voice conversations with Ray-Ban Meta, by default for training future AI models. The audio recordings of voice conversations is another matter; a person can opt out of them too. To do that, when anyone first logs into the Ray-Ban Meta app, it gives them an opportunity to decide if their voice recording can be used by training Meta's AI models.

It's quite obvious that Meta, Snap, among many other tech companies, are highly desirous of smart glasses as a new computing form factor. They all have cameras people wear on their face, and they're all essentially AI-powered. That's really raking up all of the privacy concerns we heard so long ago when we first had Google Glass. Ray-Ban Meta glasses have already been hacked by some college students who have managed to unveil the name, address, and phone number of anybody they look at, according to 404 media.