• Wearable AI requires product-level trust through deliberate opt-in controls and on-device privacy.
• Visible signals on wearable smart devices, such as indicator lights, are insufficient to protect bystanders from continuous sensing.

AI is moving beyond the screen and onto the body through devices, opening up new possibilities for everyday life, industrial work and accessibility.

This shift is happening while public trust in AI is still in flux, making it especially important to get it right. With traditional, screen-based tools, interaction is usually bounded; you open an app, type a prompt, grant permission, and close it. Body-worn systems are different. They operate in shared physical and social environments. They may also collect increasingly intimate data, influence how users think, feel and act, and observe or analyse unwitting bystanders.

This is why trust must be built into the product from the start.

Transparency for a new kind of AI interaction

The EU AI Act’s transparency obligations, which took effect in August 2026, require providers of certain AI systems to inform people when they are directly interacting with AI and set disclosure requirements for certain AI-generated content.

Meta’s Ray-Ban smart glasses faced a lawsuit over body-worn camera data after footage was reportedly routed to a Kenya-based contractor for manual review, raising questions about the product’s privacy-first positioning.

Separately, two Harvard students built a tool called I-XRAY to show how smart glasses could be paired with facial recognition to instantly reveal strangers’ names, addresses, and phone numbers – a case now recorded in the OECD’s AI Incidents Monitor.

These two examples show that the trust problem extends beyond bystanders; people may be captured by systems they never chose to engage with, while even those who voluntarily use a device may have little control over how the data collected are later processed, reviewed or shared.

Three principles for wearable AI

As wearable AI moves towards more continuous sensing, three principles matter most:

1. Design AI to support human judgement, not replace it

Wearable AI can help people make better decisions, but it should not make decisions on their behalf, especially when health, safety, work or money are involved. If a wearable device detects an irregular heart rhythm, for example, it might first alert the user and recommend seeking medical advice, while escalating further only if the user has enabled that option or the system detects clear signs of an emergency, such as loss of consciousness or a rapidly deteriorating condition.

People should be able to understand why a recommendation was made, question it, reject it, and turn off a feature or cancel a specific action at any time.

2. Build privacy into the product, not just the policy

A wearable AI device may know what a person is looking at, where they are, what they are saying and how their body is responding. That makes consent important, but consent alone is not enough. The first design question should be simpler: does the product need this data at all?

For a device that may stay with someone throughout the day, privacy begins with restraint – collect sensitive data only when needed, process it on the device whenever possible, and avoid storing raw information unless absolutely necessary.

There is also a practical incentive to address privacy early. In countries with strict privacy laws, a device may simply not be allowed to operate as it does elsewhere. If companies wait until launch to deal with these rules, they may end up rebuilding core features for individual markets. Designing for these constraints from the start can help avoid costly changes later.

3. Make sensing visible to the people around the device

The wearer may have agreed to the terms of service, but the people around them have not. In shared spaces, others may be seen, recorded, identified or analysed without their consent.

A visible signal matters, but it cannot be the whole answer. An indicator light can be missed, ignored or covered. The more sensitive the function is – identifying a face, pulling up a name or address, or analysing someone nearby – the more cautious the system should be before activating it. These capabilities should require deliberate opt-in rather than running quietly in the background.

https://www.weforum.org/stories/artificial-intelligence/8910283f-2eba-40da-bac4-7c9b963bc0d4?utm_source=sfmc&utm_medium=email&utm_campaign=2884073_ForumStories-30October2026&utm_term=&emailType=Forum%20Stories