5 of 7 Smart Glasses Brands Leave Data Retention Limits Unclear

If you ask your smart glasses to translate a conversation or to explain what you're looking at, you are in effect placing trust in a company with part of your life and possibly that of another person. But the terms of that trust are hidden within the app permissions and privacy policies and may even fail to make clear how long a recording will last. An analysis of seven smart-glasses brands carried out by Cybernews looked at this less obvious aspect of the technology (Cybernews, 2026).

Researchers used Exodus Privacy data to compare the Android apps for Meta, Rokid, RayNeo, INMO, Solos, Even Realities, and Halliday. Meta’s app asked for the most permissions: 70 in total, with 18 labeled as dangerous by Android. In this context, 'dangerous' means the app needs special approval to access sensitive information or perform certain actions (Android, 2026). This label does not mean the app has misused data. For example, Meta’s app can read SMS messages and record audio, but just having these permissions does not show when or how they are used.

Counting permissions only tells part of the story. Meta’s app had one detected tracker, called Facebook Flipper, which is a debugging tool. INMO Global and Solos AirGo each had six trackers, while Hi Rokid had none. These results show that the type of software components is just as important as how many there are.

This evidence also has limitations. Exodus Privacy states that its analysis identifies known tracker signatures in an app (Exodus, 2026). Yet the fact that a tracker is detected does not prove that the component is active, and the lack of any matches does not mean that no tracking is taking place. While a ranking may help to focus attention on particular cases, it cannot show which product deals with personal information in the most responsible way.

All seven products relied on cloud processing for their AI features. That makes the provider’s rules for handling submitted material especially important. Yet five of the seven did not specify a retention limit in the documentation reviewed. Even Realities disclosed immediate deletion by default for captured voice; Meta specified periods ranging from 30 days to a year for some data. Still, cloud processing does not automatically mean recordings are used to train AI. It does mean buyers need clear explanations of what leaves the device and how long it remains available. Without those answers, consenting to a feature becomes a decision made with incomplete information.

Bystanders have even less to say. For anyone caught in the recording, that choice may never arise. Only Meta and Rokid have confirmed safeguards that respond to a covered-recording indicator, and only three brands have offered privacy documents written specifically for their glasses.

The implication is that manufacturers are asking people to accept a socially intrusive technology without consistently explaining its consequences. A recording light can announce that recording is happening, but trust also depends on knowing where the material goes and when it disappears, all questions that matter long after the wearer has received a translation or taken a photograph.