Two Harvard students turn Meta’s smart glasses into a privacy nightmare


The students have demonstrated how private data can be exposed at a glance.

Smart glasses, due to their ability to continuously capture and share data through a video camera, pose numerous privacy concerns.

For one, they can be used for recording and surveillance without the knowledge of those around them.

ADVERTISEMENT

Combine this with AI and software that allows you to scan public databases, and you get a device that can instantly expose private information.

That’s precisely what AnhPhu Nguyen and Caine Ardayfio, two Harvard students, have created, demonstrating their findings in a video posted on X. Their project enabled them to dox strangers in real-time.

How did they do it?

For their experiment, the students used Meta Ray-Ban smart glasses, which look almost indistinguishable from regular ones. They streamed the video straight from the glasses to Instagram.

A program monitored the stream, while AI detected people when they looked at someone’s face. Camera images were processed using facial recognition models, which can match the face to publicly available images across the internet and return the URLs.

They also used publicly available data from articles and databases to learn more about the people in the images. The data included their home addresses, phone numbers, ages, and relatives from publicly available records and social media profiles.

All this information was fed back to the app on their smartphone. This allowed them to get information about their classmates and strangers on the street in real-time, highlighting the privacy issues such wearable devices pose.

ADVERTISEMENT

In an accompanying document, the students say the project, dubbed I-XRAY, aims to raise awareness that extracting a random person’s home address and other personal details from just their face is possible today.

“What makes I-XRAY unique is that it operates entirely automatically, thanks to the recent progress in large LLMs. The system leverages the ability of LLMs to understand, process, and compile vast amounts of information from diverse sources(...). This synergy between LLMs and reverse face search allows for automatic and comprehensive data extraction that was previously not possible with traditional methods alone,” they say.

For those who want to protect themselves against doxxing, students advise them to delete their data from Reverse Face Search Engines, such as Pimeyes or Facecheck ID, as well as People Search Engines, including FastPeopleSearch, CheckThem, and Instant Checkmate.