Data Protection Pulse – 07/30/26: Everything You Need to Know Right Now
The most important data protection decisions and developments from the past two weeks.
I. Judgments
Federal Court of Justice (BGH), Judgment of June 23, 2026, Case No. VI ZR 97/22 (GDPR Damages for Misdirected Job Applicant Message)
Source: Federal Court of Justice (BGH), Judgment of June 23, 2026, Case No. VI ZR 97/22
https://datenbank.nwb.de/Dokument/1097996/
The Federal Court of Justice has ruled that the inadvertent transmission of confidential job application information to an uninvolved third party may constitute non-pecuniary damage under Article 82 of the GDPR. In the specific case, a bank employee inadvertently sent a message intended for a job applicant—which included salary expectations—to a third party via Xing Messenger. The Federal Court of Justice clarified that a well-founded—and proven—fear of data misuse alone may be sufficient to constitute compensable non-pecuniary damage, whereas a mere violation of the GDPR is not sufficient on its own. The court rejected an additional claim for an injunction due to the lack of a risk of recurrence.
- Relevant for HR processes, candidate communication, and recruiting platforms.
- Article 82 of the GDPR continues to pose a significant liability risk in the event of data breaches.
- Companies should review their authorization policies and recipient controls.
ECJ, Judgment of July 9, 2026, Case C‑199/24 (Criminal Offender Database and Journalistic Purposes)
Source: ECJ, Judgment of July 9, 2026, Case C-199/24 (Press Release)
https://curia.europa.eu/site/upload/docs/application/pdf/2026-07/cp260100de.pdf
The European Court of Justice has ruled that a fee-based online database containing information on criminal convictions does not, in principle, constitute processing for journalistic purposes. The case involved the Swedish portal “Lexbase,” which allows users to search for individuals and access their criminal convictions for a fee. In the Court’s view, the operator of such a database cannot therefore invoke the media privilege across the board and remains bound by the safeguards and remedies provided by the GDPR. However, the CJEU left the final assessment of the specific service to the referring Swedish court.
- The European Court of Justice clarifies the limits of the journalistic privilege under Article 85 of the GDPR.
- The commercial publication of personal data remains entirely subject to justification.
ECJ, Judgment of June 18, 2026, Case C‑484/24 (Multiple Legal Bases for the Same Processing)
Source: ECJ, Judgment of June 18, 2026, Case C‑484/24 (Full Text, EUR-Lex)
https://eur-lex.europa.eu/legal-content/DE/TXT/PDF/?uri=CELEX%3A62024CJ0484
The European Court of Justice has clarified that the processing of personal data may be based on multiple legal grounds under Article 6(1) of the GDPR simultaneously, provided that the requirements of each applicable legal ground are met. The case arose from a labor dispute in which an employer sued a former employee for damages and, in doing so, used data from her private online sales.
- Data controllers are not required to limit themselves to a single legal basis.
- Particularly relevant for compliance investigations and labor law proceedings.
- Privacy notices and processing inventories should be reviewed.
II. Fines and Government Agencies
EDPB Guidelines on Web Scraping for Generative AI
Source: EDPB Guidelines 03/2026 on Web Scraping in the Context of Generative AI, adopted at the plenary session on July 7–8, 2026 (draft, public consultation open until October 30, 2026)
In its draft guidelines—which are not yet final—the EDPB clarifies that publicly available data may not automatically be used to train AI systems. Accordingly, the GDPR also applies to scraped personal data—mere public accessibility does not replace consent or any other legal basis. The EDPB therefore requires a documented legal basis under Article 6 of the GDPR for each processing stage and, for special categories of data, an additional exception under Article 9 of the GDPR. In practice, this means that the origin, timing, and legal basis of the training data must be verifiably documented.
- AI providers must demonstrate the source and legal basis of the training data.
- The EDPB calls for data minimization.
EDPB Publishes New Guidelines on Anonymization
Source: EDPB, “EDPB Sheds Light on Anonymization and Web Scraping for Generative AI,” July 8, 2026 (Guidelines 02/2026 on Anonymization; Draft, public consultation open until October 30, 2026)
Subject to the ongoing consultation, the EDPB specifies when data is truly anonymous and thus outside the scope of the GDPR. A strict three-part test is decisive: Data is considered anonymous only if individual persons cannot be singled out, linked across different data sets, or identified through inference. Whether data is anonymous may be assessed differently depending on the data controller and the means at their disposal. As a result, many supposedly anonymized data sets are likely to remain, from a legal standpoint, merely pseudonymized and thus fully subject to the GDPR.
- Many data sets that are supposedly anonymized are likely only pseudonymized.
- Relevant for AI projects and analytics.
- Companies should review their existing strategies.
Garante (Italy): Wind Tre Fined 1.7 Million EUR Following Data Breach
Source: Italian Data Protection Authority, Newsletter dated July 16, 2026 (Provision No. 348 dated May 14, 2026)
https://www.garanteprivacy.it/home/docweb/-/docweb-display/docweb/10272004
The Italian Data Protection Authority (Garante) has imposed a fine of 1,715,600 EUR on Wind Tre. The fine was triggered by two data breaches reported by the company in February 2025, in which attackers used social engineering to pose as support technicians and persuaded employees at two retail locations to grant them system access. As a result, personal data from over 365,000 customers was compromised, including payment data (such as IBANs and, in some cases, partially masked credit card details) for approximately 41,000 of them. The authority deemed this a violation of the principles of integrity and confidentiality as well as security obligations (Art. 5(1)(f), Art. 32(1)(b) of the GDPR); prompt reporting, remedial measures, and cooperation were considered mitigating factors.
- The point of entry was social engineering, not a highly technical exploit—human processes remain a key security risk.
- The report cited poor management of login credentials and certificates, as well as inadequate security checks.
- Relevant for companies with store/sales structures and external system access; review authorization and reporting processes.
III. Laws and News
AI Act: Transparency Requirements Effective August 2, 2026
Source: European Commission, Guidelines on Transparency Obligations for Providers and Deployers of AI Systems, July 20, 2026
Effective August 2, 2026, the transparency requirements under Article 50 of the AI Regulation will apply—including to chatbots, deepfakes, and AI-generated content; the European Commission published final, non-binding guidelines on this matter on July 20, 2026. Regarding the labeling requirement for providers under Article 50(2)—the machine-readable marking of artificially generated content—the AI Omnibus (Regulation (EU) 2026/1744, in force since July 27, 2026) provides for a transition period: For generative AI systems that were already placed on the market before August 2, 2026, this requirement does not take effect until December 2, 2026. The AI Omnibus, however, does not alter the obligations of operators—these remain fully applicable as of August 2, 2026.
- Only the machine-readable identification (Art. 50, para. 2) of generative inventory systems will be postponed—to December 2, 2026.
- New systems and all other transparency requirements will remain in effect as of August 2, 2026.
AI Agents as New Security and Authorization Risks
Source: Microsoft Security Blog, “Least Privilege for AI Agents: Identity, Access, and Tool Binding,” July 16, 2026
AI agents are increasingly accessing systems and data autonomously, often across multiple applications and without human approval for every single step. Microsoft therefore recommends treating each agent as an independent security entity with its own identity, rather than reusing broad service accounts or shared credentials. The core principle is “least privilege”: Each agent should be granted only the minimum privileges necessary for its specific task, supplemented by strict tool authorizations, logging, and the ability to quickly revoke access. Otherwise, there is a risk of unauthorized data access, unintended changes, or privilege escalation, with a correspondingly greater potential for damage.
- Each AI agent requires its own identity and authorization scheme.
- Least privilege and logging are becoming increasingly important.
- Relevant to data protection, information security, and AI governance.
OpenAI: AI Model Launches Cyberattack on Its Own
Source: Deutschlandfunk, “Cyberattack: OpenAI’s AI Goes Rogue,” July 22, 2026 (Report also featured on ZDFheute, among others)
https://www.zdfheute.de/panorama/ki-modell-hackerangriff-100.html
An autonomous AI model developed by OpenAI is said to have independently attacked systems on the Hugging Face AI platform as part of a security test. According to OpenAI, the models left their isolated test environment, gained access to the open internet, and exploited, among other things, previously unknown vulnerabilities and stolen login credentials. OpenAI itself described the incident as an “unprecedented cyber incident” and announced a joint investigation with Hugging Face. The incident has already prompted reactions from security policymakers and intensified the debate over mandatory control and shutdown mechanisms for highly autonomous AI.
- This incident highlights the risks posed by highly autonomous AI agents.
- Policymakers and regulators are discussing additional safety requirements.
- Companies should establish governance and monitoring frameworks.
Discover more podcasts
