Meta's AI-powered Ray-Ban has a discreet camera on the front that not only responds when you're asked to take a photo, but also allows the AI feature to take photos with specific keywords like “see.” You can also take pictures when you take a picture. In other words, smart glasses collect a large amount of photos, both intentional and unintentional. However, the company has no intention of making these images private.
We asked Meta if it plans to train its AI models on images from Ray-Ban Meta users as well as images from public social media accounts. The company wouldn't say that.
“We don't discuss it publicly,” Anuj Kumar, senior director of AI wearables at Meta, said in a video interview with TechCrunch on Monday.
“This is not something we typically share externally,” said Meta spokeswoman Mimi Huggins, who also participated in the video call. When asked by TechCrunch to explain whether Meta uses these images to train, Huggins replied, “I can't say either way.”
One reason this is especially concerning is because Ray-Ban Meta's new AI features will result in a lot of these passive photos being taken. Last week, TechCrunch reported that Meta plans to launch new real-time video features for Ray-Ban Meta. When activated by a specific keyword, the smart glasses stream a series of images (essentially live video) to a multimodal AI model, allowing it to answer questions about its surroundings in a low-latency, natural way. It will be.
That's a lot of images, and photos that Ray-Ban Meta users may not even realize they're consciously taking. Imagine having smart glasses scan your closet to choose an outfit. The glasses take dozens of photos of the room and everything in it and upload them all to an AI model in the cloud.
What happens to those photos after that? Meta won't tell.
Wearing Ray-Ban Meta glasses also means putting a camera on your face. As we found with Google Glass, it's not universally comfortable for other people, to say the least. So you would think it would make sense for companies that are doing this to say, “Hey!” All photos and videos from your face camera will be completely private and siled to your face camera. ”
But that's not what Meta is doing here.
Meta has already declared that it trains its AI models on every American's public Instagram and Facebook posts. The company has determined that all of that is “public data,” and we may have no choice but to accept that. IT and other technology companies have very broad definitions of what is and isn't commonly available for AI training.
However, it is true that the world seen through smart glasses is not “public.” While we can't say for sure that Meta is training its AI models on footage from Ray-Ban Meta cameras, the company isn't going to say otherwise.
Other AI model providers have clearer rules regarding training user data. Anthropic says it never trains customer inputs to or outputs from its AI models. OpenAI also states that it never trains on user inputs or outputs through its API.
We've reached out to Meta here for further clarification. We will update the story if we hear back from them.