Ray-Ban Meta AI Glasses – Analysis under the EU GDPR and the UK GDPR – Part 1

This informal CPD article ‘Ray-Ban Meta AI Glasses – Analysis under the EU GDPR and the UK GDPR – Part 1’ was provided by Educage Training, a team of legal and technical professionals specializing in data protection, information security, and regulatory compliance.

According to the Hamburg Commissioner for Data Protection and Freedom of Information (HmbBfDI), recordings made with Ray-Ban Meta AI glasses of people outside a close circle of friends and family will generally not be lawful from a data protection perspective, apart from rare cases that can be based on legitimate interests, because informed consent cannot be obtained in practice and there is usually no legal basis for AI training by Meta¹,².

The analysis reaches three main conclusions:

  1. The wearer is a controller. A private individual using the glasses also falls within the scope of the GDPR; the household exemption must be interpreted narrowly and ceases to apply if the recording is made public or reaches Meta for AI training².
  2. Transparency is the weak point. The design of the glasses conceals the fact that they are a recording device, and the small LED does not provide reliable notice from a distance, in sunlight or when Meta AI is used².
  3. AI training is the most critical processing operation. If the wearer has not objected to training, they become a joint controller with Meta and would have to demonstrate a dual legal basis, which, according to the report, is usually not possible in the case of passers-by².

The relevant provisions of the UK GDPR follow essentially the same logic, but the UK Information Commissioner's Office (ICO) has not yet published an analysis on this issue comparable in detail to the Hamburg report: according to publicly available information, it has asked Meta for an explanation³,⁴, and there is no specific legislation on smart glasses⁵.  The Hamburg position is not binding on the ICO, but its line of reasoning is also useful for a UK assessment. 

At the end of the analysis, the problems identified and the available options are summarised separately in Part 2.

On 13 October 2026, the European Data Protection Board (EDPB) adopted a statement on smart glasses (Statement on smart glasses)¹⁹. The statement is not detailed guidance; it sets out the EDPB's preliminary reflections, but it confirms the main points of the Hamburg report at EU level, as follows:

  • the household exemption is narrow, 
  • with current indicator lights, passers-by typically do not know that processing is taking place, 
  • consent is difficult to obtain from non-users, 
  • and legitimate interests are particularly difficult to justify for AI training¹⁹. 

Going beyond the Hamburg report, it also addresses the responsibility of the manufacturer and of the provider of the related (AI) service¹⁹.

A sociological study was also published in October 2026 under the EDPB's Support Pool of Experts programme (Dr Yann Bruna, Université Paris-Nanterre), which examined the social reception of the glasses on the basis of 22 interviews²⁰. The report does not reflect the EDPB's official position, but it clearly shows that some users use the glasses to gather evidence, and many interpret the LED as a sign of implied consent²⁰.

1. Background and methodology of the review

The Hamburg authority examined the “Ray-Ban Meta Wayfarer (Gen 1)” model as a case study, with a team of experts in social networks, technical data protection and the legal aspects of AI applications¹. The report was published on 10 September 2026². The glasses contain a 12-megapixel camera, five microphones, two speakers, 32 GB of flash memory, Wi-Fi and Bluetooth, and are controlled via the Meta AI mobile app². The technical review consisted of four elements:

  1. an analysis of the app's structure and permissions, 
  2. capture of the network traffic, 
  3. an examination of the app's local storage, and 
  4. disassembly of the glasses. 

The team also submitted a data request via Meta's Accounts Centre². Most of the network traffic remained encrypted, and the memory chip of the glasses could not be read². The legal analysis focuses exclusively on the responsibility of the wearer as controller in relation to the data of third parties (passers-by, onlookers, companions), along three usage scenarios: 

  1. photo and video recording,
  2. use of Meta AI, and 
  3. sharing on social networks². 

The report does not assess Meta's own processing².

The report itself acknowledges that it is not exhaustive, is based on the first-generation model, has not examined professional use (other than by content creators) in detail, and can only partly be generalised to the general issues raised by smart glasses². This analysis relies on the English version of the report; the section on social networks could only be partly reviewed because the retrieved text was truncated, so for that section I rely on the report's summary.

The analysis also takes into account two new documents connected to the EDPB: the EDPB statement of 13 October 2026¹⁹ and the findings of a sociological report commissioned by the EDPB on the proposal of the CNIL²⁰. The latter was carried out between March and August 2026 through semi-structured interviews with 7 users and 15 non-users, and expressly reflects the author's views, not the EDPB's official position²⁰.

2. Technical findings underpinning the legal assessment

Recording indicator LED: The outward-facing LED is clearly lit only when taking photos, recording video, live streaming and making video calls. When Meta AI is used – for example, when the wearer asks about their surroundings or the people they can see – it is only faintly lit on the Gen 1 model and not lit at all on Gen 2². The light is emitted at a narrow angle and is barely visible outdoors in sunlight. The examiners were able to cover the LED with a sticker, paint or a cap in such a way that enough ambient light still reached the light sensor for the camera to work, while little or no light was visible from the outside. For video recording, the system checks whether the LED is covered only at the start, so an LED taped over during recording has no effect².

The EDPB statement also states more generally that, given the current state of the art, the indicator lights on smart glasses leave many passers-by unaware. A white LED can easily be mistaken for a reflection or solar glare, and people with visual impairments cannot perceive it at all¹⁹. 

The interviewees in the sociological report had the same experience. The LED cannot be seen in sunlight, it does not show where the data goes (local storage or live stream), and it can be circumvented with accessories available from online shops or through software²⁰.

Database tables pointing to facial recognition: The app's database contains tables named “face”, “face_group” and similar. These were empty at the time of the review, so, according to the report, no actual facial recognition was taking place, but the structure of the software would already be capable of it². The report refers to reports by the EFF and WIRED from June 2026, according to which matching faces against a database could be made to work through manual intervention, after which Meta modified the software; according to the EFF's update of 8 June 2026, Meta removed the inactive facial recognition code from a newer version of the app²,¹⁷.

Face masking: It could not be established whether faces in images sent to Meta AI are masked before transmission. The timestamp of the masked images showed a time after transmission, and because the traffic was encrypted the question could not be resolved². In the data export requested via the Accounts Centre, faces appeared masked with a grey circle².

Voice recordings: During setup, the system asks whether voice commands may be stored; Meta intends to use these for future AI training, and in this respect the European version differs from the US version. Withdrawing consent does not retroactively delete earlier recordings; it only stops new ones from being saved². To delete the chat history, the entries in the “Voice Activity Log” have to be deleted separately, in a different part of the app².

Other observations: The “Hey Meta” wake word was recognised reliably and only in the wearer's voice. The app requests wide-ranging permissions (contacts, call log, SMS, precise location), and it would not actually need to request gallery permission in order to import images. The data export also contained hardware, connection and usage logs, as well as imprecise location data².

3. Assessment under the EU GDPR

3.1 Scope and the household exemption

Article 2(2)(c) and Recital 18 of the GDPR exclude from the Regulation's scope processing carried out in the course of a purely personal or household activity. Relying on the case law of the Court of Justice of the European Union (Lindqvist, Ryneš, Buivids) and EDPB practice, the Hamburg authority interprets this exemption narrowly²,¹³,¹⁴. 

The factors to be weighed are: the objectively ascertainable purpose of the processing, publicity (making data accessible to an indefinite number of people), the location, the adverse impact on data subjects, and the overall picture². The exemption may apply, for example, to a video of a family holiday kept as a private memento, but not to a recording shared publicly on social media². The exemption also ceases to apply if the wearer has not exercised the right to object to AI training, because in that case third parties' data end up in a non-private training environment determined by Meta². 

The exemption does not apply at all to professional users (influencers, content creators). The media privilege (Article 85) can only come into play for journalistic activity involving editorial control and serving a debate of public interest, not for every piece of uploaded content²,¹³.

The EDPB statement confirms this narrow interpretation at EU level. The exemption may apply if the wearer shares the photo or video only with a family member or friend, or uses the translation function for their own purposes in a private setting. It does not apply, however, if the recording reaches an indefinite number of people via a public social media account, or if the use takes place in the course of official, professional, political or other non-personal activities¹⁹. In such cases the wearer is a controller or joint controller and is subject to all the obligations of the GDPR, including transparency¹⁹.

The EDPB highlights two further points. The household exemption applies only to the wearer: the GDPR still applies to the manufacturer and to the provider of the related app or (AI) service, even if the wearer uses the device for private purposes (Recital 18)¹⁹. In addition, a wearer making audio or video recordings may also be subject to other laws, such as criminal law provisions¹⁹.

3.2 Responsibility and joint controllership

The wearer decides when and where the camera is pointed, what is recorded and what Meta AI is asked to do, and is therefore an independent controller within the meaning of Article 4(7)². If AI training is switched on, the report, relying on the case law of the CJEU (Wirtschaftsakademie, Fashion ID), finds joint controllership with Meta under Article 26, at least as regards the collection and transmission of the data and their transfer into the training environment²,¹³. Meta does not currently offer such an arrangement for Meta AI, and joint controllership does not remove the need for a legal basis: the wearer must demonstrate a legal basis both for their own processing and for the disclosure to Meta².

The EDPB statement also addresses the question that the Hamburg report deliberately left out: the role of the manufacturer and the service provider. The manufacturer of the glasses, or the provider of the related app or AI service, may be an independent or joint controller if it determines the purposes and means of the processing alone or jointly with others. According to the EDPB's example, this is the case where the provider of the AI function analyses the camera image – in which other people are also visible – in order to give the wearer an answer about their surroundings¹⁹.

Where the manufacturer or the service provider is a controller, it cannot shift the fulfilment of its own obligations onto the wearer, including, for example, informing passers-by¹⁹. If it also uses the data for a new purpose, it must demonstrate that the new purpose is compatible with the original one (Article 5(1)(b), Article 6(4)), paying particular attention to the fact that some data subjects have no relationship with it whatsoever¹⁹.

cpd-Educage-Training-CJEU-judgment-on-body-cameras
CJEU judgment on body cameras

3.3 Transparency and data protection by design

The glasses follow the classic Wayfarer shape, with the camera and the LED concealed by decorative metal elements, so passers-by do not expect a smart device capable of recording². According to the report, the flashing white LED does not provide timely and adequate notice that processing has begun, which is contrary to Article 5(1)(a) and Articles 12–14². The assessment builds on the CJEU judgment on body cameras (C-422/24, 18 December 2025) and the German DSK guidance on body cameras: the most important information must be made available immediately when recording begins (layered information), and portable cameras require continuous, clearly visible signalling and verbal notice, or a high-visibility vest with lettering²,¹³,¹⁵. The shortcoming primarily concerns the manufacturer's obligation of data protection by design and by default (Article 25), but according to the report the wearer, as controller, must also make up for it².

According to the EDPB statement, transparency is one of the biggest challenges posed by smart glasses. A traditional privacy notice is not sufficient to inform non-users, particularly where the manufacturer or service provider processes their data, since it has no relationship with them and cannot reach them¹⁹. The EDPB therefore recommends that manufacturers combine several mechanisms and signal processing by default and in real time; these signals should be impossible or difficult to switch off, and easy for passers-by to notice in practice¹⁹. 

Examples given in the statement:

  • Light signal (LED): it should be visible in a wide range of lighting conditions, stay lit long enough, work differently depending on the function (for example, a continuous light when recording and a flash when taking a photo), and it should not be possible to record when the LED is covered¹⁹.
  • Audio signal: for example, a shutter sound when taking a photo, or activation tied to a spoken command¹⁹.
  • Smartphone proximity notification: apps already exist that detect nearby glasses on the basis of their Bluetooth identifiers, and manufacturers could also send a specific signal while recording. However, this cannot be the only mechanism, receiving notifications should be optional, and the app must not lead to the tracking of wearers¹⁹.
  • Design: the shape of the frame could also help distinguish smart glasses from ordinary ones¹⁹.

The EDPB stresses that where transparency relies on technical solutions, these must be effective, difficult to circumvent and regularly updated, and it recommends the development of industry standards and awareness-raising campaigns¹⁹. The sociological report provides practical background to this: some critical non-users would expect a conspicuous red indicator light that cannot be covered, and a design that departs from the traditional frame shape²⁰.

Under data protection by design and by default in Article 25, the EDPB also recommends further technical measures: limiting continuous recording, live streaming and retention periods; blurring passers-by's faces, pseudonymisation or anonymisation before sharing or re-use, preferably on the device; for devices designed to provide real-time assistance, disabling persistent storage of recordings by default; location-based restrictions (geofencing) in sensitive places; and “privacy frictions” that slow down sharing on social media¹⁹. The statement also indicates that in certain cases the glasses may qualify as terminal equipment, so the ePrivacy Directive may also apply, particularly where the device's sensors collect, store and process image, video or audio material locally, or make it available to a remote server by default¹⁹.

3.4 Legal bases in three usage scenarios

Photo and video recording

  • Consent (Article 6(1)(a)): Possible in principle, but notice must be given before recording; the LED alone is not enough².
  • Legitimate interests (Article 6(1)(f)): In most cases, the three-step test comes out in favour of passers-by unless the wearer takes additional measures².

Meta AI without AI training

  • Consent (Article 6(1)(a)): Possible in controlled, pre-arranged situations; practically impossible in everyday situations².
  • Legitimate interests (Article 6(1)(f)): Sometimes yes: navigation, translation, information requests, social participation of people with visual impairments².

Meta AI with AI training

  • Consent (Article 6(1)(a)): Practically ruled out in everyday situations (lack of information and of the ability to withdraw)².
  • Legitimate interests (Article 6(1)(f)): Usually fails the balancing test; for special category data, Article 9 is also an obstacle².

Legitimate interests require three cumulative steps: a legitimate interest, necessity, and a balancing test against the interests of the data subject, which must be documented (EDPB Guidelines 1/2024, CJEU C-621/22)²,¹³,¹⁴. Reasonable expectations are particularly important: a passer-by in a public space can expect that they may be recorded by a mobile phone, but not that their face will be transmitted to an AI provider by the camera in an ordinary-looking spectacle frame². For Meta AI use, the necessity test also asks whether the purpose can be achieved in a less intrusive way (for example, by turning the camera away or leaving out personal data)².

AI training: In its decision of 23 May 2025, the Higher Regional Court of Cologne accepted in principle that Meta may rely on legitimate interests to use public adult user content and Meta AI interactions for training, but according to the report this situation is only partly comparable to the glasses: there, the platform relationship, deliberately published content and an effective means of objecting are missing²,¹⁶. According to EDPB Opinion 28/2024, personal data in training data may persist in the model's parameters and may be extractable, which weighs against the data subject in the legitimate interests balancing test²,¹⁴. Article 9(2)(e) (data manifestly made public by the data subject) cannot be applied to third parties². The withdrawability of consent (Article 7(3)) also raises unanswered questions for training that has already taken place².

On legal bases, the EDPB statement reaches a conclusion consistent with the Hamburg report¹⁹. The validity of a legal basis requires a case-by-case assessment, taking into account local laws, the specific processing and any further use of the data; for special category data, one of the exceptions in Article 9(2) must also apply¹⁹. 

Consent may be an option where the conditions of Article 7 are met; where national law protects the right to one's image, it is likely to be the most appropriate legal basis¹⁹. The manufacturer or service provider can hardly obtain freely given, specific, informed and unambiguous consent from non-users, and consent requested afterwards – for example, after the recording – is invalid, because the legal basis must exist before and during the processing¹⁹.

According to the EDPB, legitimate interests require a careful, case-by-case assessment. The current lack of transparency undermines passers-by's reasonable expectations, so a fair balance cannot be struck without appropriate mitigating measures¹⁹. This is particularly difficult to justify for further processing for commercial purposes beyond the service expressly requested by the wearer, and for AI training, which passers-by do not expect¹⁹. The EDPB also states that meeting the transparency requirements does not in itself mean that the data subject can reasonably expect the processing, and that where legitimate interests are relied on, every data subject must be able to object at any time (Article 21(1)), which is often difficult in practice¹⁹.

3.5 Special category data and vulnerable data subjects

Recordings and voices may reveal special category data (Article 9), such as health, religious or political beliefs. In certain circumstances, a voice may also qualify as biometric data. Facial recognition – if the function were switched on – would be processing of biometric data for the purpose of unique identification, subject to the strict conditions of Article 9². The report specifically highlights children, very elderly people, residents of women's refuges, people attending funerals, and sensitive locations (doctor's surgeries, hospitals, places of worship, nurseries, schools, saunas)².

The EDPB statement lists similar sensitive situations: healthcare facilities, schools and other places frequented by minors, changing rooms, swimming pools and beaches, toilets, and other private or intimate situations; in these, recording is inappropriate or may be prohibited by national law or local rules¹⁹. 

It calls for particular attention to the protection of children, as they are less likely to recognise the presence of the glasses and their implications¹⁹. The EDPB considers use in combination with facial or voice recognition to pose a high risk to rights and freedoms in public spaces¹⁹.

According to the EDPB, use in the workplace also raises concerns regarding employee monitoring and the protection of trade secrets and intellectual property¹⁹. The sociological report supports this with specific cases: a public-sector manager reported that a colleague had secretly recorded strategic meetings with the glasses, which led staff to censor themselves²⁰. Another interviewee wondered where their health data had gone after an acquaintance had Meta AI “read out” their blood test results²⁰.

3.6 Sharing on social media

Public posting generally requires separate consent that also covers publication, and can be based on legitimate interests only within narrow limits. In the legitimate interests balancing test, the Hamburg authority uses the case law under the German Art Copyright Act (KUG) as a benchmark, which is a German specificity: in other Member States, local rules protecting a person's image provide similar guidance².

According to the EDPB statement, sharing via a public social media account falls outside the household exemption, and to mitigate the risks it recommends “privacy frictions” that slow down sharing, and blurring passers-by's faces before sharing¹⁹. According to the sociological report, sharing is at the centre of social concerns: interviewees fear that a few seconds of footage taken out of context could lead to online “lynching”²⁰.

The content of this article is intended solely for general information and educational purposes and does not constitute legal advice or a legal opinion in any form. The content of the article is based on publicly available sources and is not intended to assess specific legal situations or individual cases.

The author and the publisher strive to ensure that the content is accurate and up to date, but accept no liability for any omissions, outdated information, or any direct or indirect loss arising from the use of the content. The relevant legislative framework and the related regulatory and judicial practice are subject to change, so the article cannot be regarded as an exhaustive or up-to-date summary of the legislation in force at any given time. 

We hope this article was helpful. For more information from Educage Training, please visit their CPD Member Directory page. Alternatively, you can go to the CPD Industry Hubs for more articles, courses and events relevant to your Continuing Professional Development requirements.

REFERENCES

  1. HmbBfDI: Ray-Ban Meta AI Glasses – HmbBfDI legt technischen und datenschutzrechtlichen Prüfbericht vor, 10 September 2026
  2. HmbBfDI: Final Report on the Technical and Data Protection Review of the Ray-Ban Meta AI Glasses
  3. The Register: UK watchdog eyes Meta's smart glasses after workers say they 'see everything'
  4. Business & Human Rights Resource Centre (Chris Vallance, BBC News, 4 March 2026): UK data watchdog requets proof that Meta smart glasses are compliant with data protection law
  5. Chambers Student (legal careers website, Erin Bradbury, 27 July 2026): The Memo: Keeping one eye on smart glasses and the other on growing privacy concerns
  6. IFSEC Insider: ICO publishes new guidance on domestic CCTV following multiple cases of misuse
  7. Information Rights and Wrongs (Jon Baines's blog, 8 July 2023): ICO guidance on domestic CCTV – more hindrance than help
  8. TechCrunch: Brits Using Google Glass Warned To Heed Data Protection Law (27 June 2014, from the time of the pre-GDPR data protection law)
  9. Mayer Brown: The Data (Use and Access) Act: Reform of the GDPR in the UK (1 July 2025)
  10. Clyde & Co: The Data (Use and Access) Act 2025 (July 2025)
  11. GOV.UK: Data (Use and Access) Act 2025: data protection and privacy changes
  12. Tees: The Data (Use and Access) Act 2025: what this means for UK Businesses
  13. Judgments of the Court of Justice of the European Union, as cited in source 2 (HmbBfDI report): C-101/01 Lindqvist; C-212/13 Ryneš; C-345/17 Buivids; C-210/16 Wirtschaftsakademie; C-25/17 Jehovan todistajat; C-40/17 Fashion ID; C-621/22 Koninklijke Nederlandse Lawn Tennisbond; C-422/24 AB Storstockholms Lokaltrafik (18 December 2025); C-199/24 Legal Newsdesk Sweden (9 July 2026)
  14. EDPB documents, as cited in source 2: Guidelines 3/2019 on processing of personal data through video devices; Guidelines 1/2024 on processing of personal data based on legitimate interests; Opinion 28/2024 on certain data protection aspects related to AI models
  15. DSK (Conference of the German data protection authorities): Orientierungshilfe zum Einsatz von Bodycams durch private Sicherheitsunternehmen (22 February 2019)
  16. OLG Köln, decision of 23 May 2025, 15 UKl 2/25 (Meta AI training)
  17. EFF: Move Fast, Surveil Things  and Meta Silently Added Face-Recognition Code for Its Smart Glasses to Millions of Phones 
  18. Information Rights and Wrongs (Jon Baines's blog, 18 July 2026): The extent of the purely personal and household carve-out - Phillips v Oloffs (2026) 
  19. EDPB: Statement on Smart glasses (adopted on 13 October 2026)
  20. Dr Yann Bruna (Université Paris-Nanterre): The connected gaze. Video recording, surveillance and administering evidence in the age of smart glasses (October 2026)