CHI 2026

See Me If You Can: A Multi-Layer Protocol for Bystander Privacy with Consent-Based Restoration

Yahya Khawaja*, Shirin Rehman*, Alexander Ponticello, Divyanshu Bhardwaj, Katharina Krombholz, Muhammad Hamad Alizai, Naveed Anwar Bhatti* Equal contribution

In short

Camera glasses now look like ordinary eyewear, so people nearby often do not notice that they are being filmed. Most existing protections expect bystanders to opt out. This paper builds a privacy-by-default alternative, in which faces stay hidden unless the person agrees, and asks wearers and bystanders what they think of it.

How the study was done

  • 18 interviews
  • 9 wearers, 9 bystanders
  • 3-tier protocol

We designed and implemented a three-tier protocol: faces are blurred on the device at capture, can be replaced with synthetic faces, and are restored only after the person consents. 18 participants, 9 wearers and 9 bystanders, then used the protocol interface and discussed it in semi-structured interviews.

Key findings

  1. Bystanders

    Opt-in protection is essential

    Bystanders saw mandatory, privacy-by-default protection as essential and asked for even stronger anonymization.

  2. Consent

    Asking depends on the relationship

    Whether wearers sent a consent request, and whether bystanders accepted it, depended on who they were to each other and on the situation.

  3. Wearers

    Safe, but limiting

    Wearers valued the safeguards but found the protected recordings visually limiting and the consent step cumbersome.

  4. Context

    Protection that adapts to the situation

    Wearers wanted to apply and relax the protocol depending on context, for example with pre-approved bystanders.

What should change

Privacy-enhancing technologies for camera glasses should adapt to context and give wearers and bystanders an effective way to negotiate consent. The implementation shows that an opt-in, privacy-by-default design is technically feasible on wearable-class hardware, and the paper gives directions for camera glass manufacturers.

Common questions

Can smart glasses record people without their consent?

Yes. Camera glasses now look like ordinary eyewear, so people nearby often do not notice that they are being filmed, and most existing protections expect bystanders to opt out. This paper tests the opposite: faces stay hidden unless the person agrees.

How does the opt-in privacy protocol work?

It has three tiers: faces are blurred on the device at capture, can be replaced with synthetic faces, and are restored only after the person consents. The implementation shows this is technically feasible on wearable-class hardware.

What did wearers and bystanders think of it?

Bystanders saw mandatory, privacy-by-default protection as essential and asked for even stronger anonymization. Wearers valued the safeguards but found the protected recordings visually limiting and the consent step cumbersome, and wanted to adapt the protocol to the situation, for example for pre-approved bystanders.

Abstract

The growing popularity of wearable camera glasses raises pressing concerns about bystanders being recorded without their consent. Most existing privacy-enhancing technologies (PETs) rely on opt-out models that place the burden of privacy protection on bystanders. We conducted a qualitative study on wearers’ and bystanders’ perceptions of opt-in, privacy-by-default approaches for camera glasses. To enable this study, we designed and evaluated an opt-in privacy-by-default protocol. We then conducted semi-structured interviews with camera glass wearers and bystanders (N = 18) to examine their perceptions of the protocol. Our findings show that bystanders viewed the opt-in protocol as essential and advocated for even stronger anonymization. Wearers appreciated the protocol’s safeguards but found it visually limiting, expressing desire for a context-dependent version that can be enabled in relevant scenarios. Our findings highlight the need for context-aware PETs that provide effective mechanisms for consent negotiation.

Cite this paper