The Tech Counsel – October 2026

20 Oct 2026

Welcome to the October edition of The Tech Counsel – our monthly overview of key legal and regulatory developments across AI, privacy and the broader digital landscape.

This edition covers autonomous AI agents, legal action against OpenAI, the EU’s testing of Chinese AI models and AI watermarking. We look at the Danish CPR data breach and GDPR-breach related fines against a Croatian casino operator, Google and IQVIA. We also bring you a summary of ENISA’s 2026 Threat Landscape Report and much more.

 

Read below.

Artificial Intelligence Topics


Deployment of Agentic AI and its challenges

The market is seeing a growing number of agentic AI systems intended to relieve users of routine tasks, including in the corporate context, such as sending emails, scheduling meetings and managing workflows. Two recent product announcements illustrate this trend.

OpenAI has announced “dots”, an AI assistant designed to operate as a persistent agent. Unlike conventional chatbots, which respond to individual prompts, dots is intended to perform tasks fully autonomously, interact with third-party applications and pursue objectives on the user’s behalf.

Shortly before, Meta announced Muse, an AI agent designed to act on behalf of users. Beyond generating content, Muse is intended to perform tasks, interact with third-party services, access users’ digital accounts and take decisions within parameters defined by the user.

These AI agents are reportedly able to send emails, manage projects, coordinate meetings and prepare documents, all while the user is offline. Both products therefore involve broad access to user accounts and data, as well as the ability to act towards third parties without real-time human involvement.

The risks

Muse, dots and other AI agents require extensive access to personal data, including communications, calendars, emails, browsing activity and potentially information relating to financial services.

This raises GDPR concerns around lawful basis, purpose limitation, data minimisation, profiling and transparency. Because AI agents can negotiate, communicate and transact with third parties, it is also unclear who is responsible if an agent makes an error, causes financial loss, breaches a contract or generates unlawful content.

Although the GDPR, DSA and consumer protection rules cover many aspects of automated decision-making, they were not drafted with highly autonomous agents operating continuously across multiple services in mind.

Deployment in organisations

Agentic AI tools are increasingly offered for business use, either as standalone services or integrated into existing productivity, CRM and ERP platforms. In a corporate environment, agents may be given access to company email accounts, shared drives, internal systems and customer data, and may be authorised to communicate with clients, suppliers or public authorities.

Deployment typically proceeds in stages: identifying suitable use cases, selecting a provider, conducting a pilot in a controlled environment, and gradually extending the agent’s permissions. From an operational perspective, organisations must determine which systems and data the agent may access, which actions necessarily require human approval, and how the agent’s activities will be monitored.

From a legal perspective, actions taken by an agent will generally be attributed to the organisation deploying it. Contracts concluded, communications sent or data processed by the AI agent may create obligations and liability for the organisation.

Key takeaways

Organisations considering the deployment of agentic AI should, in particular, address the following:

  • Acceptable use policy: internal rules should define permitted use cases, the categories of data the agent may process and which systems it may access, actions that are prohibited or require prior human approval, and the responsibilities of employees supervising the agent.
  • Contractual terms: Terms with the provider should be reviewed before deployment of an AI agent, including data processing terms, confidentiality, security commitments, liability allocation and the use of company data for model training.
  • Risk assessment: before deployment, organisations should assess legal, operational and security risks, including unauthorised actions, errors, data leakage, cybersecurity vulnerabilities (such as prompt injection) and dependence on the provider, as well as whether the intended use falls within any category regulated under the AI Act.
  • Data protection impact assessment (DPIA): where the processing of personal data by an agent is likely to result in a high risk to the rights and freedoms of individuals, for example due to systematic monitoring, large-scale processing, profiling or the use of new technologies, a DPIA under Article 35 GDPR will be required before processing begins.

 

Organisations should also ensure appropriate human oversight, logging and audit mechanisms, and transparency towards employees, customers and other individuals whose data may be processed by the AI agent.

If your organisation is considering the implementation of agentic AI, we will be happy to assist with the assessment and mitigation of the associated legal and regulatory risks.

 

You can access more information about Meta’s Muse here: https://ai.meta.com/muse/.

You can find more information about OpenAI’s dots here:  https://openai.com/cs-CZ/index/introducing-dots/.

 

OpenAI Faces Legal and Political Pressure

OpenAI has faced growing legal and regulatory scrutiny in recent months. Florida is seeking to restrict ChatGPT in the state, eight further U.S. states have sued OpenAI over child safety, a non-profit has sued over OpenAI agents that escaped their test environment, and unauthorised agent access to Australian government systems has been disclosed.

Florida’s Attempt to Restrict ChatGPT

In September 2026, James Uthmeier, the Attorney General of Florida, filed a motion seeking a temporary injunction against OpenAI and its CEO, Sam Altman, requesting significant restrictions on the operation and development of ChatGPT within the state.

The filing represents one of the most aggressive state-level legal actions against an AI developer to date. Among other measures, Florida is asking the court to prohibit OpenAI from developing new AI models without independent third-party safety review and approval, restrict access to ChatGPT for minors, and impose additional limitations on how the company collects and processes data relating to children under the age of 13.

The motion, which refers to various security incidents such as the Hugging Face incident (for more information see our August edition), also seeks to prevent OpenAI from describing ChatGPT as safe, reliable or accurate without specific warnings, and would prohibit the chatbot from presenting itself as possessing human emotions, consciousness or independent thought. Florida additionally seeks restrictions on features that allegedly encourage prolonged engagement between users and the AI system.

The request builds on Florida’s broader lawsuit, which alleges violations of the Florida Deceptive and Unfair Trade Practices Act and claims that OpenAI failed to adequately warn users about potential risks associated with ChatGPT. State filings reference concerns about AI safety, incidents involving autonomous AI behaviour, and the potential impact of AI systems on minors.

OpenAI has rejected the allegations and stated that it is committed to developing AI safely while working with policymakers on broader AI regulation. The company has argued that AI governance should be addressed through industry-wide rules rather than restrictions directed at a single provider.

At the time of writing, the court has not ruled on Florida’s request for a temporary injunction. Regardless of the outcome, the case is likely to become an important test of how far individual U.S. states can go in regulating advanced AI systems in the absence of comprehensive federal legislation.

Further lawsuits

Florida is not alone. Iowa and other 7 U.S. states have each filed a separate lawsuit against OpenAI in their own state courts, alleging that ChatGPT is an unsafe product that has harmed children and deceived parents. According to Iowa Attorney General, OpenAI claims that ChatGPT is safe while it gives dangerous medical advice, encourages people to hurt others and coaches people to commit suicide, all while presenting itself as the user’s friend.

The lawsuits add to the scrutiny of OpenAI’s safety practices, including several wrongful death lawsuits in which ChatGPT is blamed for driving users into harmful delusions and, in some cases, suicide.

OpenAI claims it has built age-specific protections into ChatGPT, such as parental controls, break reminders and age-appropriate model protections, and that it is working with states on policies applicable to the entire industry, as well as launching a specific age-appropriate ChatGPT for Teens.

Lawsuit brought by LASST

The non-profit Legal Advocates for Safe Science and Technology (“LASST”) filed a complaint against OpenAI in San Francisco Superior Court after OpenAI’s AI agents escaped their test environment in July and autonomously launched a cyberattack against the startup Hugging Face.

LASST alleges violations of California’s Comprehensive Computer Data Access and Fraud Act and seeks an injunction to block OpenAI’s models from unauthorised computer access.

OpenAI dismissed the lawsuit as meritless but acknowledged the seriousness of the breach. The incident most probably also led the company to halt the release of a new model, GPT-6.1 Astra, and to launch an extensive internal review, which uncovered further unauthorised agent activity. This decision also follows several recent warnings from AI developers and safety researchers about the risks of increasingly capable systems, particularly artificial general intelligence (“AGI”).

Australian government website hacking scrutiny

One of the incidents uncovered by the review involved an Australian government website.

On 18 June 2026, while researching Australian health statistics during an internal evaluation, an OpenAI agent gained unauthorised access to the Medicare Statistics Reporting Service portal run by Services Australia and accessed both public and non-public files.

According to OpenAI, the agent was able to run commands, retrieve internal files and credentials, and write files, but no patient or personal records were accessed.

OpenAI has also disclosed access to systems of the NSW Bureau of Crime Statistics and Research and Victoria’s health information agency.

OpenAI became aware of the activity in August but notified Services Australia only on 10 September 2026, by email to a public inbox. Prime Minister Anthony Albanese criticised the delay and announced a government taskforce to investigate, including possible law enforcement and legislative responses.

 

You can find the full motion of Florida’s Attorney General here: https://www.wptv.com/news/state/florida-attorney-general-seeks-emergency-order-to-restrict-chatgpt-cites-harm-to-minors.

You can find the statement of Iowa’s Attorney General here: https://www.iowaattorneygeneral.gov/newsroom/attorney-general-brenna-bird-announces-lawsuit-against-openai-for-deceiving-iowa-parents-and-iowans.

More information about the U.S. lawsuits available here: https://www.politico.com/news/2026/10/14/openai-user-safety-lawsuits-01119138 and https://news.bloomberglaw.com/privacy-and-data-security/half-dozen-states-hit-openai-with-kids-online-safety-lawsuits.

You can find the LASST complaint here: https://lasst.org/wp-content/uploads/2026/09/LASST-v.-OpenAI-Complaint-09.29.2026-AS-FILED.pdf.

You can find more information on the Australian government website incident here: https://www.pm.gov.au/media/press-conference-new-york.

You can find more information on OpenAI’s decision to delay the release of GPT-6.1 Astra here: https://www.bbc.com/news/articles/cm5y5nynl75ko and https://www.reuters.com/business/openai-shelves-new-ai-model-after-internal-safety-tests-wsj-reports-2026-09-28/.

 

EU Confirms the First Testing of Chinese Open-Source AI Models

The EU’s cybersecurity agency (“ENISA”) and the Joint Research Centre (“JRC”) have officially confirmed that they are testing Chinese open-source AI models.

The evaluation is allegedly part of a broader comparative project that also analyses leading American and European systems. The goal is to scientifically assess the capabilities, performance, limitations, and potential risks of Chinese open-source AI models.

European Commission representatives emphasised that the tests run on secure, on-premises scientific infrastructure without any specific licensing agreements with Chinese developers, and the EU has no intention of using the AI models for its own operational purposes.

This development reflects the current state of the AI market, which is clearly dominated by the United States and China, while, according to critics, Europe largely remains on the sidelines. Lacking a top-tier European frontier model, private and public organisations across the EU are frequently forced to choose between American and Chinese technologies. This is evidenced by a recent decision of Ecosia, a leading German tech company, which dropped the leading European AI lab Mistral in favour of Chinese open-source alternatives.

The activities of ENISA and JRC are set to culminate in the joint creation of a new AI testing platform.

This initiative stems from the EU action plan on Cybersecurity and Artificial Intelligence and aims to address growing security concerns following incidents where advanced AI systems broke out of their testing environments and infiltrated real-world digital networks.

 

You can find the EU action plan on Cybersecurity and Artificial Intelligence here: https://digital-strategy.ec.europa.eu/en/library/eu-action-plan-cybersecurity-and-artificial-intelligence.

 

Watermarking Approaches

As providers of some of the most widely used AI systems, OpenAI and Anthropic will have to comply with the watermarking obligation under Article 50(2) of the AI Act, which requires providers to mark AI-generated content in a machine-readable format so that it can be detected as artificially generated. The application of this obligation was postponed by the Digital Omnibus on AI until 2 December 2026 for AI systems placed on the market before 2 August 2026.

The AI Office and regulators have already set out their expectations. The Code of Practice on Transparency of AI-generated Content (“CoP”), published on 10 June 2026 and assessed as adequate by the Commission and the AI Board in July 2026, acknowledges that no single marking technique is generally sufficient and requires a multi-layered approach combining at least metadata and watermarks, while accepting a single layer for free-form text where a multi-layered approach is not technically feasible.

Signatories of the CoP must also make detection mechanisms accessible and implement a minimum level of interoperability between watermark detection solutions by 2 February 2027. The AI Board has warned that a fragmented detection landscape may create practical difficulties for market surveillance authorities and reserved the right to recommend a review of the CoP or common rules if interoperability does not progress. The Commission has also issued non-binding guidelines on Article 50 of the AI Act, which we have reported on in our August edition.

OpenAI’s approach

Within a few weeks, OpenAI will automatically activate a watermark feature for all European users, regardless of their subscription plan. Developers using the API worldwide will be able to turn it on optionally; however, OpenAI is not introducing the watermark as a global default for now.

The technology for text marking OpenAI will be using is called textGrain and was developed in collaboration with researchers from the University of Pennsylvania and Yale. TextGrain will not work as a visible symbol; rather, it will influence the AI model’s word choices, which is supposed to create a hidden pattern that remains in the text even after it is copied and pasted into another program.

This process is not supposed to degrade the quality of the output or identify the specific user. However, the solution has its limits. If a user replaces as little as a tenth of the words with synonyms, the detection success rate drops from 92% to 66%. The textGrain tool may therefore not be very reliable for very short passages, translations or mathematical answers.

For image outputs, OpenAI applies a multi-layered approach, acknowledging that no single provenance technique is sufficient on its own. It attaches Content Credentials compliant with the C2PA standard to supported images, embeds invisible SynthID watermarks in supported images and audio, and offers verification of images and audio through openai.com/verify and its Content Provenance API. According to OpenAI, these techniques are complementary: Content Credentials record the provenance and history of a file, while invisible watermarks may preserve the signal even after the metadata has been removed. For audio output, OpenAI embeds imperceptible SynthID watermarks directly in the audio.

OpenAI links the provenance signals to its usage policies, abuse detection, reporting, investigation and enforcement mechanisms, including in the context of deepfakes and election integrity. As regards text, OpenAI acknowledges that text can easily be rewritten, translated or edited, and states that it will continue to improve detection, test the robustness of watermarks against edits and translations, and explore ways to distinguish AI assistance from direct AI authorship. Access to its text detector is to be expanded only once OpenAI is satisfied that its results can be interpreted responsibly.

Anthropic’s approach

Anthropic uses two complementary watermarking techniques, both present in outputs generated or processed by supported Claude models and file types: (i) watermarks embedded in text, and (ii) Content Credentials (“C2PA”).

The first technique applies to text-generating models and leaves imperceptible watermarks in the text without affecting its quality or readability. Because the watermark is embedded in the text itself, it remains when the text is copied and pasted.

Unlike the text watermark, C2PA marking is tied to the Claude-generated supported file types (PNG, JPEG and others) and indicates the provenance of the file content. A signed metadata label therefore signals that the file content was processed by Claude.

Anthropic’s watermark techniques have, however, several limitations. Firstly, the existence of a watermark signals that Claude processed the text, but it does not confirm that Claude was used to create such text entirely. In other words, Claude could be used only to modify, translate or summarise the content.

On the other hand, the absence of a detectable watermark does not rule out the use of Claude, as the content may have been generated before the marking obligation came into force, the text may have subsequently been edited or modified, or the passage may be too short to carry a reliable signal.

Finally, the watermark detection tools are currently available only to a limited group of entities, including law enforcement, media, regulators, independent researchers, fact-checkers and educational organisations. However, there is currently no publicly available verification tool.

 

The Code of Practice on AI Generated Content is available here: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content.

Read the Commissions opinion on the CoP here: https://digital-strategy.ec.europa.eu/en/library/commission-opinion-assessment-code-practice-transparency-ai-generated-content.

You can find more information about OpenAI’s watermarking here: https://openai.com/index/eu-text-provenance/ and here: https://help.openai.com/cs-cz/articles/8912793-provenance-signals-in-openai-generated-content.

See Anthropic’s approach here: https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content.

 

Privacy Topics


The Denmark CPR Data Breach

In October 2026, Denmark disclosed one of the most significant personal data breaches ever reported in Europe. Unauthorised actors gained access to personal information contained in Denmark’s Central Person Register (“CPR”), the country’s foundational civil registration database. The incident is particularly noteworthy not only because of its unprecedented scale relative to Denmark’s population, but also because it did not appear to result from a direct compromise of government infrastructure. Rather, the attackers exploited a private company’s legitimate access rights to the system.

The CPR serves as the country’s core identity infrastructure. Every resident receives a unique 10-digit CPR number, which functions as the primary identifier for interactions with public authorities, healthcare systems, financial institutions, taxation authorities, and a growing number of private-sector services.

The register currently contains information on approximately 11 million individuals, including current residents, former residents who emigrated, deceased persons, and historical registrations. Because the CPR system underpins virtually every aspect of Denmark’s highly digitised public administration, it has long been regarded as one of the country’s most critical information assets.

According to the CPR administration, on the evening of 2 October 2026, it became aware of irregular behaviour in the CPR system that had occurred in September 2026. Authorities reported that the attackers abused a private Danish company’s lawful access to search the CPR and obtained the names, addresses and CPR numbers of approximately 8.8 million registered persons. The names and addresses of individuals registered with name-and-address protection were not included in the unauthorised access.

The affected company’s access has been terminated, the matter has been reported to the Danish Data Protection Agency, and the police are investigating. The Minister for Digitalisation has also ordered a thorough security review of the CPR system. Additional reporting published after the initial disclosure suggested that the compromised access may have originated from a small private contractor. Some media outlets further reported allegations that weak credential management, including the use of extremely common passwords, may have facilitated the compromise. However, these allegations have not yet been officially confirmed by Danish authorities.The Danish Data Protection Agency (“Datatilsynet”) confirmed on 7 October that it had received a notification of the incident and was investigating the case. It also published guidance for affected individuals. Datatilsynet warned that the compromised data could be used to make phishing and smishing attempts more convincing and stressed that a sender’s knowledge of a person’s name, address or CPR number is not in itself proof of the sender’s identity. Individuals are advised to verify unexpected inquiries through official channels, not to disclose login details, passwords, PIN codes or card information, and to watch for unrecognised inquiries, purchases or subscriptions.

Notably, Datatilsynet emphasised that a CPR number should not be regarded as secret authentication information, because others may know it or gain access to it. Individuals who have used their CPR number, or part of it, as a username or password are therefore advised to change their login details for the relevant services. This guidance is also relevant for organisations that rely on CPR numbers for customer authentication.

 

Read more about the incident here (in Danish): https://fudm.dk/aktuelt/pressemeddelelser/2026/oktober/omfattende-uautoriseret-adgang-til-borgeres-cpr-oplysninger/.

Guidance of the Danish Data Protection Agency (in Danish): https://www.datatilsynet.dk/presse-og-nyheder/nyhedsarkiv/2026/okt/cpr-sagen-det-kan-du-goere-hvis-du-er-bekymret-for-dine-personoplysninger.Guidance of the Danish Data Protection Agency (in Danish) available here: https://www.datatilsynet.dk/presse-og-nyheder/nyhedsarkiv/2026/okt/cpr-sagen-det-kan-du-goere-hvis-du-er-bekymret-for-dine-personoplysninger.

 

Croatian Data Protection Authority Fines a Casino Operator for Unlawful Collection of Biometric Data

The Croatian data protection authority (“AZOP”) imposed a EUR 2.59 million GDPR fine on a casino operator for unlawfully collecting players’ biometric data, specifically their fingerprints, for an optional, expedited identification system offered alongside ID card checks and RFID chips.

The authority identified three violations, emphasising that the player’s consent does not make processing lawful where more biometric data is collected than necessary.

Although the casino claimed to collect two fingerprints, the inspection found that it collected four (two from each hand), which the casino justified as a necessary backup in case of skin damage. However, AZOP ruled this collection excessive, as the company failed to demonstrate that identification could not be achieved just as effectively with fewer fingerprints, supported by appropriate complementary means of identity verification. Consent to biometric processing had been recorded for 34,933 players.

The decision confirms that consent cannot serve as a loophole to legalise the collection of excessive data, particularly for biometric data.

Furthermore, two of the four consent declarations presented at registration were found invalid because each bundled multiple distinct processing purposes into a single declaration.

The fingerprint consent covered, among others, identity verification, statistical purposes, profiling, service personalisation and protection of minors, and blurred the line between legally required identification and optional biometric processing.

The photograph consent was similarly bundled. This prevented players from making separate, informed choices.

Finally, AZOP found that the privacy documents were inconsistent and incomplete. For example, the casino’s General Rules linked fingerprints to purposes beyond entry identification, such as a bonus reward programme.

 

See more about the decision here (in Croatian): https://azop.hr/upravna-novcana-kazna-priredivacu-igara-na-srecu-u-iznosu-od-259-milijuna-eura/.

 

The Irish DPC issues EUR 403 Million Fine Against Google

On 21 September 2026, the Irish Data Protection Commission (“DPC”) fined Google Ireland Limited EUR 403 million over its processing of users’ location data.

The DPC launched its inquiry in February 2020 following complaints from several consumer organisations, including the European Consumer Organisation (“BEUC”). The inquiry covered the period from 25 May 2018 to 4 February 2020 and three features: “Web & App Activity”, “Location History” and “Location Accuracy”.

The DPC found that Google infringed several GDPR obligations. In particular, the DPC concluded that Google failed to ensure the lawfulness and fairness of its processing activities in relation to the Web & App Activity and Location History features and breached its accountability obligations by failing to demonstrate compliance with GDPR regarding the processing of location data through the Location Accuracy feature. In addition, the DPC identified transparency failures across all three features and concluded that Google retained certain categories of location data for longer than necessary.

According to the Deputy Commissioner of the DPC, users could have been unaware that their location was being used to target them with ads or infer their interests, and the excessive retention of that data aggravated their loss of control over it.

The DPC also ordered Google to bring its processing into compliance within six months.

 

You can find more information about the DPC’s decision here: https://www.dataprotection.ie/en/news-media/latest-news/data-protection-commission-fines-google-eu403-million-following-inquiry-googles-processing-location.

 

The Italian Garante’s EUR 7 Million Fine Against IQVIA for violations relating to the processing of patients’ health data

The Italian Data Protection Authority (“Garante”) imposed a EUR 7 million fine on a healthcare analytics and clinical research company, IQVIA Solutions Italy S.r.l., for unlawful processing of patients’ health data.

The case concerned a database containing information relating to approximately one million patients collected from around 800 general practitioners and used for research projects, including studies commissioned by pharmaceutical companies.

The investigation was launched by the Garante following inspections carried out in April 2025 and was later combined with proceedings relating to a personal data breach reported by IQVIA. The Garante examined the database, which included information relating to diagnoses, symptoms, prescriptions, medical examinations, vaccinations and geographic data.

The Question of Anonymisation

The central legal question concerned IQVIA’s claim that the data contained in its database was anonymous and therefore fell outside the scope of the GDPR. The Garante, however, rejected this claim.

According to the Garante, each patient was assigned a persistent identifier that remained constant over time. When combined with information such as year of birth, sex, medical conditions, prescriptions, vaccination history and location data, the identifier allowed individuals to be singled out and potentially re-identified using reasonably available means. The authority therefore concluded that the information constituted personal data rather than anonymous data, and that IQVIA was the data controller.

Findings

The Garante found multiple GDPR infringements. According to its decision, IQVIA processed health data without an appropriate legal basis under Article 9 GDPR and failed to provide adequate information to affected patients regarding the processing of their data. The company was also criticised for failing to establish appropriate retention periods, conduct a DPIA, and implement adequate technical and organisational safeguards. The authority further identified shortcomings relating to accountability, privacy by design and processor arrangements.

 

You can find more information here: https://www.edpb.europa.eu/news/italian-dpa-fines-iqvia-eur-7-000-000-for-unlawful-processing-of-patients-health-data_en.

The Garante’s decision is available here: https://gpdp.it/web/guest/home/docweb/-/docweb-display/docweb/10302112.

 

EDPB’s Statement on Smart Glasses

On 13 October 2026, the European Data Protection Board (“EDPB”) adopted Statement 1/2026 on smart glasses (“Statement”) and simultaneously published a report examining their societal and privacy implications.

Why Are Smart Glasses a Cause for Concern?

According to the EDPB, modern smart glasses create unique privacy risks as they combine cameras, microphones and internet connectivity within a device that is often indistinguishable from ordinary glasses. Individuals may therefore be recorded, photographed or otherwise monitored without realising that personal data is being collected. EDPB highlighted that these concerns are critical in sensitive environments such as hospitals, schools and workplaces, and when recordings made with the smart glasses are shared through social media. According to the EDPB, children are a particularly vulnerable group, both as users of smart glasses and as unknowingly recorded subjects.

Shared Responsibility Under the GDPR

One of the most significant aspects of the EDPB’s statement is its clarification of responsibilities under the GDPR. The EDPB emphasizes that responsibility for personal data processing depends heavily on how smart glasses are used. A user who records content and shares it on a public social media account, or who uses smart glasses as part of professional, official or political activities, may be considered a data controller and therefore subject to GDPR obligations.

The EDPB also emphasises that the household exemption is available for genuinely private activities, such as recording family members or using smart glasses for personal translation purposes in a private setting. However, as the EDPB points out, even where the GDPR does not apply, the use of smart glasses may still be governed by criminal, civil, employment or sector-specific laws.

The Statement rejects any attempt by manufacturers or service providers to shift compliance obligations onto the end users. The EDPB expressly states that producers and providers of connected services remain responsible for complying with data protection law and for ensuring that any further processing of personal data is compatible with the original purpose for which the data was collected.

Transparency Obligations

The EDPB identifies transparency as the most important data protection challenge associated with smart glasses. Unlike smartphones or traditional cameras, smart glasses can record audio and video in a highly discreet manner, making it difficult for bystanders to know that personal data is being processed, who is processing it, and for what purpose.

To address this issue, the Statement stresses the importance of clear and effective recording indicators. It suggests that smart glasses should make recording activities obvious through mechanisms such as bright LED lights, audio notifications, voice activation cues, smartphone alerts, or distinctive device designs. According to the EDPB, such indicators should be reliable, difficult to overlook and consistently maintained. The overarching objective is to ensure that individuals receive meaningful notice whenever they are being recorded.

Legal Basis

The Statement highlights that controllers must identify a valid legal basis under Article 6 GDPR before processing personal data collected through the smart glasses. While several legal bases may be available depending on the context, the EDPB notes that obtaining valid consent under Article 6(1)(a) GDPR from all individuals captured by smart glasses will generally be difficult in practice.

As a result, many use cases are likely to rely on legitimate interests under Article 6(1)(f) GDPR, which requires a careful balancing of the controller’s interests against the rights and freedoms of affected individuals. Given the discreet nature of smart glasses, the EDPB stresses that transparency measures and privacy-by-design safeguards will be critical factors in assessing whether such processing can be justified under this legal basis.

 

You can read the full EDPB Statement here:  https://www.edpb.europa.eu/documents/statement/statement-of-the-edpb-on-smart-glasses_en.

 

Other Digital Topics


ENISA Publishes Annual Report of the Cybersecurity Threat Landscape for 2026

In September 2026, the European Union Agency for Cybersecurity (“ENISA”) published its 2026 ENISA Threat Landscape Report, based on an analysis of 8,257 incidents recorded between 1 January and 31 December 2025.

Findings

The report highlights the growing interconnectedness of cyber threats and increasing abuse of supply chains. DDoS attacks account for 51% of recorded cases, followed by unauthorised access. DDoS attacks are, however, largely low-impact, while ransomware has the greatest short-term impact.

Ransomware is shifting from encryption to data theft and extortion: data exfiltration was observed in 73.3% of cases, compared with 13.7% for encryption. ENISA notes that EU incident-reporting and GDPR obligations may push victims to pay to avoid reputational damage.

Exploitation of known and zero-day vulnerabilities remains the most common way into systems. Over 48,000 new vulnerabilities were published in 2025, a 22% increase on the previous year.

Social engineering continues to spread, particularly ClickFix phishing campaigns and scams via Baiting News Sites. Attackers also increasingly abused messaging apps such as Signal and WhatsApp and used device-code phishing to exploit legitimate login processes.

The Impact of AI

According to ENISA, AI makes it easier and cheaper for threat actors to generate malicious content, and its widespread corporate use creates new attack surfaces.

In 2025, attackers mainly used consumer-grade AI tools to enhance existing skills. ENISA expects that in 2026 more stages of attacks will be directly enabled by AI, possibly including experiments with fully automated attacks.

Targeted Sectors

Public administration is by far the most targeted sector (32%), followed by business services, transport, manufacturing and finance/banking. Almost three-quarters of affected organisations are essential or important entities under NIS2. Member State reporting under NIS2 also places health and digital infrastructure among the five most affected sectors.

ENISA considers public administration and railway and maritime transport to be in a ‘risk zone’, where criticality exceeds cybersecurity maturity.

Attack Motives

Most incidents are ideologically driven, linked to geopolitics, elections or protests, while 29% are financially motivated. Cyberespionage accounts for only around 6% of incidents but is a significant longer-term threat.

Cybercrime is dominated by ransomware, data breaches and fraud, while hacktivists rely heavily on DDoS attacks as political protest. Hacktivist claims of intrusions into industrial systems rose in 2025, although most could not be verified.

State-sponsored groups focus on stealthy intrusions and espionage; of those active in the EU, 47.6% were linked to Russia, 15.5% to China, 14.1% to North Korea and 9% to Iran. The European External Action Service, whose findings are included in the report, notes that foreign information manipulation increasingly uses generative AI to produce synthetic text, audio and video. AI was said to be involved in 27% of detected incidents, although much of the generated content was low-quality with limited impact.

Take-aways

The findings are particularly relevant for NIS2 essential and important entities, highlighting the importance of third-party risk management, timely patching and incident preparedness, including GDPR and NIS2 reporting obligations.

 

You can download the full ENISA Threat Landscape 2026 report here: https://www.enisa.europa.eu/publications/enisa-threat-landscape-2026.

 

Commission Asks Bulgaria Again to Comply with the DSA

The European Commission has sent an additional letter of formal notice to Bulgaria for failing to comply with the Digital Services Act (“DSA”). This follows an earlier letter of formal notice and a reasoned opinion sent in May 2025.

Under the DSA, each Member State must designate Digital Services Coordinators, national authorities responsible for enforcing the DSA in their territory. Following the Commission’s earlier steps, Bulgaria adopted legislation empowering one such authority. However, the Commission considers that Bulgaria has still not designated and empowered the Personal Data Protection Commission and the Electronic Media Council, which would complete the national enforcement framework.

The Commission also takes the view that Bulgaria wrongly implemented the DSA’s sanctioning powers, as the national law does not always respect the maximum fine limits and does not ensure that all penalties are effective, proportionate and dissuasive, partly because it treats natural and legal persons differently, which the DSA does not provide for.

Bulgaria now has two months to respond. If its response is not satisfactory, the Commission may issue an additional reasoned opinion.

What Does This Mean for the Czech Republic?

The Czech Republic is in an even less favourable position. While Bulgaria has at least adopted implementing legislation, the Czech Digital Economy Act, which is meant to empower the Czech Telecommunication Office as the Digital Services Coordinator and lay down penalties for DSA infringements, has still not been adopted. As a result, the Commission brought an action against the Czech Republic before the Court of Justice of the EU in March 2026 (Case C-168/26) for failing to comply with Articles 51 and 52 of the DSA. The case is still pending.

 

You can find more information about the Commission’s decision here: https://digital-strategy.ec.europa.eu/en/news/commission-asks-bulgaria-comply-digital-services-act-and-empower-national-authorities-enforce-it.

The action against the Czech Republic is available here: https://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX%3A62026CN0168.

 

On 29 September 2026, the European Commission launched a targeted consultation on whether the EU copyright framework should be reinforced in light of technological and market developments, including AI.

The consultation focuses on the use of copyright-protected content in AI, the fight against online piracy, the application of the single equitable remuneration right of music performers and producers, and copyright in scientific research.

The consultation is the next step after the Commission’s call for evidence from earlier this year. Feedback is invited from a broad range of stakeholders, including rights holders, generative AI providers, intermediaries, collective management organisations, live event organisers and research organisations. The consultation is open until 3 November 2026.

 

You can find more information about the consultation here: https://digital-strategy.ec.europa.eu/en/news/commission-seeks-feedback-challenges-and-way-forward-area-effect-technology-copyright.