Meta will stop training its AI on ‘visual data’ from its smart glasses — if you opt out
Meta has introduced an opt-out feature allowing users to prevent visual data from their smart glasses being used to train the company's artificial intelligence models. The move comes in response to growing privacy concerns about the company's use of camera-based interactions for AI training, addressing at least one proposed class-action lawsuit. This represents a significant concession from the technology company, which faces ongoing backlash over privacy practices surrounding its AI-equipped eyewear.
Previously, Meta allowed images and videos captured during Meta AI interactions—such as asking the glasses to translate a sign—to be used for training purposes, though many users were unaware these captures were being stored at all. The company had outsourced the review and labelling of these images to third-party contractors in other countries, some of whom reportedly encountered intimate and personal imagery during this work. Under the new opt-out setting, users who decline will have their visual data deleted immediately after answering their query, with neither training nor product review occurring. However, the opt-out only applies to visual data; audio interactions with Meta AI remain subject to the same training and review practices, though users can manually delete audio recordings.
- Meta now lets users opt out of AI training on smart glasses visual data
- Previously, contractors reviewed personal images without users' knowledge
- Audio data still used for training; camera-less glasses alternative launched
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Allowing users to opt out of visual-data training gives people meaningful control over intimate images and videos that may be captured without their full awareness. Immediate deletion after a query can reduce privacy risks, limit exposure to contractors and address legitimate concerns about consent and surveillance. Supporters may also see the change as a reasonable accountability measure for a company that benefits from users’ data.
The case against
Meta could argue that using visual interactions to train and review its models improves translation, image recognition and other features for everyone. An opt-out system may reduce the data available for testing and refinement, potentially slowing progress or making services less reliable, while still leaving users free to delete recordings or decline participation. The company might also say that applying different rules to visual and audio data reflects differing technical and operational needs rather than disregard for privacy.