I advise on artificial intelligence projects, in particular on the Data Act, the AI Act and the Digital Services Act, and prepare data protection impact assessments under Art. 35 GDPR.
Classifying AI systems under the AI Act
Data protection impact assessments under Art. 35 GDPR
The OECD - OCDE’s AI Incidents and Hazards Monitor tracks what goas wrong with AI. Its data comes from Event Registry, a platform processing more than 150,000 news articles daily …
A team wants to use an AI tool to analyse customer messages, summarise HR documents or prepare decisions. Before discussing the model, clarify the processing: 1) What personal …
On 2 November 2026, German Federal Government published its target vision "Germany as an AI Nation 2030", setting out 18 cross-government measures for the years ahead. Some of the …
This sentence should trigger a closer review. Employees may enter customer information, internal documents, contracts or personal data into an AI service. If the provider uses …
Imagine an AI service reliably taking over work that previously cost a company €100,000 a year HR costs. At €75,000, it could still look attractive, provided the costs of …
A German municipal swimming pool operator has suspended the introduction of an AI-based camera system. The technology is intended to detect swimmers at risk of drowning and alert …
Looking through the draft Certified AI Specialist Legal (AISL) body of knowledge, I can see why the two could work well together. On research, it distinguishes between fabricated …
OpenAI’s Astra system card (Article 55 AI Act requires providers of GPAI models with systemic risk to evaluate models, mitigate systemic risks, report serious incidents and ensure …
I think we need to change the way we think about AI governance. Think of agents as a digital workforce. An agent should have an identity. It should have a role and job …
In 1949, J. Robert Oppenheimer chaired the committee that advised the US Government against an all-out programme to develop the hydrogen bomb. Having helped open the atomic age …
On 12 September 2026, the EU Data Act will have applied for one year. For manufacturers of connected products and providers of related services an important design obligation …
I increasingly see articles saying that lawyers can use AI services such as Claude, provided they comply with professional secrecy, data protection and appropriate security …
In The Organizational Behavior of Agentic AI (July 2026), Canhui Liu describes agentic AI as a "partial organisational analogue". Agent collectives can divide labour, coordinate …
This is a timely reason to examine a practical question: which AI-generated or manipulated content must actually be labelled? For images, audio and video, the relevant obligation …
Google published a White Paper on A Pragmatic Approach to AI Governance in America (June 2026). Google argues for a differentiated governance model based on risk and capability …
Its central point is that data quality and the GDPR accuracy principle are closely connected, but not interchangeable. Data quality is broader, applies to personal and …
The European Commission’s AI Act Service Desk has provided one of the clearest statements so far on how AI agents fit within the AI Act framework. The key takeaway is that AI …
The Blue Guide explains how core EU product law concepts work in practice. Things like CE marking, conformity assessment, technical documentation, market surveillance, and the …
If an organization is just beginning to set up AI governance, what's the most important first step toward an effective AI risk management system without disrupting daily business? …
Since 25 March 2026, Microsoft says that Flex routing is turned on by default for eligible new EU and EFTA tenants. During peak demand, this can mean that Copilot inferencing is …
The European Commission has made these icons available for creators, publishers and other deployers of generative AI systems as part of Section 2 of the Code of Practice on …
For anyone working with generative AI, this matters. The code is designed to support compliance with the AI Act transparency obligations under Article 50, which become applicable …
The publication of the first roadmaps is an important step. The planned online consultation is equally valuable because it opens the process to wider input. Experts, practitioners …
The proposal looks at the underlying infrastructure that will determine Europe’s ability to build, deploy and scale AI in practice: cloud services, computing capacity, data …
Two are already in force: the prohibition of certain AI practices since February 2025, and the rules for general-purpose AI models since August 2025. The remaining four apply to …
On 8 May 2026, the European Commission published draft guidelines on the transparency obligations in Article 50 AI Act and opened a targeted stakeholder consultation. These …
Late last night, the Council and the European Parliament reached a provisional agreement on a targeted AI Act Omnibus package, as part of the EU’s broader regulatory …
Video‑based content dominates adolescents’ media use. At the same time, deepfakes and video‑based disinformation are becoming more realistic, faster to spread, and harder to …
I asked Deathbyclaud whether my European tech‑law firm still has a future. The verdict? “SWEATING — about to be replaced by a 1.3 KB markdown file.” A bit rude. Apparently, AI …
On 23 February 2026, the European Data Protection Board (EDPB) joined 60 other authorities in issuing a Joint Statement on AI‑Generated Imagery and the Protection of Privacy …
Over the past months, the discussion around effective anonymisation of personal data for use in Large Language Models (LLMs) has intensified. The central conclusion is clear …
AI systems are not only more popular; they are getting better or at least more fluent at handling requests and interacting with people. Controllers and supervisory authorities are …
One of the key concepts introduced is the provider: any organisation that develops an AI system or model, or places it on the market or into service under its own name or …
Now applicable since 12 Sep 2025. “Access‑by‑design” for new connected products applies to items placed on the market after 12 Sep 2026. Two pathways to IoT data Art. 3: by …
The European Data Protection Supervisor (EDPS) has released its TechSonar Report 2025–2026, focusing on emerging technologies and their implications for privacy and fundamental …
An expert in Germany who used AI extensively in preparing a court report without disclosing it saw his fee reduced to €0. Similar legal consequences should apply in other …
Certificates have traditionally signalled effort and competence. But when AI agents can earn them faster than humans, without any cognitive engagement, their significance erodes …
AI and algorithmic systems are no longer just tools that support business processes. In many organizations, they already shape, steer and evaluate human work. According to recent …
The German Data Protection Authorities (DSK) published an orientation paper in October 2025 on Retrieval-Augmented Generation (RAG) in generative AI systems. RAG enriches prompts …
AI-powered browsers promise smarter, faster browsing: summarising articles, automating tasks, even booking trips. But they also introduce new attack surfaces. One major …
With Copilot Actions, the assistant doesn’t just suggest, it acts. Booking a hotel, sending an email with local attachments, deleting duplicate files – all triggered by a prompt …
AI systems improve, self-update, meaning risk levels evolve over time. Compliance cannot be a one-off exercise. It requires continuous, real-time monitoring and in-life change …
OpenAI and Apollo Research took this question seriously. Their empirical study stress-tests deliberative alignment as an anti-scheming intervention. The idea: train models to …
At WWDC 2025, Apple unveiled new AI features like live translations in Messages, FaceTime, and calls, plus Visual Intelligence for screenshots and deeper integration of Apple …
Measuring AI governance requires structured KPIs, continuous monitoring, and alignment with recognized frameworks like ISO/IEC 42001 and NIST AI RMF. Research offers valuable …
The European Commission has initiated a public consultation to develop guidelines and a Code of Practice on transparency obligations for AI systems, as mandated by the AI Act. The …
Infosys’ latest research, based on a survey of 1,500 senior executives across North America, Western Europe, and ANZ, reveals that 95 % of companies have experienced negative …
A colleague recently pointed out that while I’ve been deeply focused on AI and data privacy, the practical side of technology implementation is just as critical. That feedback …
The German Federal Office for Information Security (BSI) and France’s ANSSI have just released updated guidance on the secure integration of Large Language Models (LLMs). Their …
Google’s latest model, Gemini 2.5 Deep Think, introduces a new reasoning mode designed for deeper, more deliberate responses. But how does it perform on safety and …
The Federal Digital Minister recently introduced the "AI Service Desk" by the Federal Network Agency, aiming to support small and medium-sized enterprises (SMEs) in AI …
"Artificial Intelligence in the Military Domain and its Implications for International Peace and Security: An Evidence-Based Road Map for Future Policy Action," is produced by …
Generative AI is no longer just about generating text or images but it's evolving into a powerful tool for innovation in drug discovery, engineering, education, architecture, and …
According to Senator Cantwell, previous amendments from Senators Marsha Blackburn and Ted Cruz would have hindered states' abilities to safeguard children and consumers. With the …
Here is the system card for Claude Opus 4 and Claude Sonnet 4 from Anthropic. Highlights: 1) A wide range of pre-deployment safety tests aligned with their Responsible Scaling …
The Conference of Independent Data Protection Supervisory Authorities of the Federal and State Governments in Germany has published the "Guidelines on Recommended Technical and …
AI video tools like Google’s Veo 3 and Runway can now generate scenes that look almost indistinguishable from reality. WSJ’s Joanna Stern and Jarrard Cole put them to the test in …
The EU’s Artificial Intelligence Act, published in July 2024, takes a risk- and capability-based approach to regulating AI systems and GPAI models like large language models …
The EU AI Act (Art. 4 and Art. 29–31) requires transparency and risk mitigation for deployed AI systems, especially those interacting with minors. Doesn’t that imply we also need …
Today the Higher Regional Court of Cologne/Germany rejected an emergency motion by Verbraucherzentrale NRW aimed at blocking Meta (Facebook, Instagram) from using publicly shared …
The GPT-4 System Card presents a structured analysis of the model’s capabilities, limitations, and emergent risks, including this warning: "This document contains content that …
According to a new Agent Governance field guide, AI agents are systems that autonomously achieve goals in the real world, with minimal human instruction. Think: not just …
For anyone working in AI governance, risk, or security these resources are essential for understanding how to evaluate and secure AI/ML systems. Testing & Evaluation / Threat …
Led by Stanford’s Center for Research on Foundation Models (CRFM), HELM (Holistic Evaluation of Language Models) benchmarks 30+ of the most prominent LMs (OpenAI, Anthropic, Meta …
The attached paper is a valuable resource for anyone interested in the ethical implications of AI and the social dynamics at play. I encourage you to read it - if not yet known …
For all who’ve lost track of the AI standards, frameworks, and certifications: you’re not alone. Three key documents are central to how organizations structure their approach …
A collaboration between Simon Mylius (Winter Fellow at the Centre for the Governance of AI) and the AI Incident Database (AIID) has resulted in a promising proof-of-concept …
The Center for Security and Emerging Technology (CSET) has developed a detailed framework to help policymakers, developers, and researchers think systematically about AI-related …
As AI tools become increasingly sophisticated, the potential for candidates - especially adversaries - to simulate human-like responses raises critical ethical and practical …
The GDPR imposes strict requirements for processing sensitive information, but there are practical approaches to ensure compliance while harnessing the power of AI. Here are some …
I recently tried a popular AI prompt: “Based on what you know about me, generate an image of your guess of what I look like. Don't worry about flattering me.” ChatGPT did not …
With AI enhancing military operations across cyber, land, air, naval, and space domains, it’s essential to understand both its potential and implications. Here are key insights …
EU digital legislation is becoming increasingly complex. The AI Regulation, which came into force last year, is intertwined with the GDPR. However, how these laws interact is not …
While the EDPB's Opinion is a step in the right direction. However, the emphasis on detailed compliance measures and the need for thorough due diligence reflects the increasing …
I recently came across this fascinating video showcasing an autonomous Tesla reacting swiftly in a critical situation, potentially saving a life. In the clip, a man falls onto the …
This directive acknowledges the role of AI in shaping product liability laws and applies to AI systems. Here are the key points you need to know: 1. AI providers (AI Act), will be …
This initial version, prepared by Chairs and Vice-Chairs of four thematic working groups, sets the foundation for guiding the development and deployment of trustworthy and safe …
ONKIDA is a fantastic aid for understanding the main principles and rights and how they are covered by CNIL, EDPS, and German authorities. 📑 Key Features of ONKIDA: - Central …
The German Federal Office for Information Security (Bundesamt für Sicherheit in der Informationstechnik or BSI) and the French National Cybersecurity Agency (Agence nationale de …
AI-driven regulatory compliance automation is transforming how businesses navigate the labyrinth of complex regulations across various sectors. With machine learning and natural …
In September 2022, the European Commission proposed a directive on AI civil liability (AILD) with an impact assessment. The European Parliament's Committee on Legal Affairs (JURI) …
The Bavarian Data Protection Authority (BayLDA) published comprehensive guidelines titled “Start-up aid from Ansbach on AI & data protection” to foster the data …
Are you a practitioner looking to conduct a GDPR data protection impact assessment on an AI system? Understanding the risks involved is crucial for compliance and ethical AI …
The study by Idoko et al. (2024) sheds light on these pressing issue. Key insights from my point of view: - The presentation of case studies and expert analyses provided valuable …
Recently, I explored Character.ai, a platform where you can chat with digital personas of both fictional characters and real people. It’s a fun and fascinating tool, but it also …
As fascinating as these innovations are, they bring with them critical discussions about privacy and IT security. Imagine a future where these robots aren't just cleaning but also …
Generative AI tools have the power to transform various facets of your business, but navigating GDPR compliance when leveraging existing data for AI training is crucial. Here's a …
The Hamburg Labour Court recently made a noteworthy ruling on the co-determination of ChatGPT and AI systems (Decision of 16.01.2024 - 24 BVGa 1/24). Here's what happened …
Thankfully, the document contains some clear statements I miss in similar publications: "Data protection’s accuracy principle applies to all personal data, whether it is …
Contrary to misconceptions, GDPR does not inhibit AI innovation in Europe. It does, however, require a meticulous approach when dealing with datasets containing personal data. The …
Do you think the EU will be able to catch up with China and the US in the GenAI patent race? I have my reasonable doubts. Instead of trying to catch up it could be beneficial to …
The 1st meeting of the European Artificial Intelligence Board (formed by Member State representatives) took place in Brussels on June 19th. The AI Board delved into the …
The EU has taken a step forward. The Council of the European Union has amended the EuroHPC regulation, expanding its mandate to include 'AI factories', entities that provide AI …
The development of AI brings immense potential but also raises important ethical and safety concerns. The Alan Turing Institute's guide, "Understanding Artificial Intelligence …
1️⃣ Retention of Training Data: The recommendations emphasize that retaining training data can enable audits and bias measurement. However, this should be limited to necessary …
The EU GDPR's transparency requirement mandates that organizations must be clear about how they collect, store, and use personal data. The requirements are relatively …
In "Machine Unlearning" Chandrasekaran et al. (2020) shed light on innovative machine unlearning techniques like sharded, isolated, sliced, and aggregated training. These …
The European Court of Auditors' Special Report 08/2024 has shone a light on the EU's AI ambition and the need for robust governance and focused investment going forward. The …
The gap between the U.S. and Europe in AI innovation is staggering and growing rapidly. According to the Stanford University AI Index, the U.S. is pulling far ahead in the AI …
Critics like Wallace and Castro (2018) argue that GDPR's rigorous requirements for user consent and its restrictions on data collection could slow down Europe's AI advancement …
Erasing personal information from AI systems isn't as straightforward as deleting a file from your computer. The intricacies of these systems and the ways data intertwines within …
The GDPR has set a high bar for data protection, introducing the "Right to be Forgotten." But how can we ensure compliance in the context of advanced AI models? Machine unlearning …
The Bavarian State Commissioner for Data Protection (BayLfD) has shared crucial insights in its brief informative note (Issue No. 55) concerning the relationship between the …
As someone involved in the field of technology, it's a privilege to witness such groundbreaking advancements. The potential of these AI-generated videos is immense, with …
Recently, a fascinating paper was published by Munkhdalai, Manaal Faruqui, and Siddharth Gopal from Google, titled "Leave No Context Behind: Efficient Infinite Context …
France’s data protection authority, CNIL, has released its first set of recommendations for using AI technologies while ensuring the protection of personal data, a crucial …
The German Federal Office for Information Security has published an enlightening document on generative AI models, with a focus on Large Language Models (LLM). These AI models are …
Selecting the right AI detector tool can be daunting with numerous options available. It's important to consider your specific needs: are you focused on individual analysis or …
What if I told you that the Data Act has the potential to redefine the way we interact with IoT devices and their data? The Act establishes that the user, whether a buyer or …
A research group at Cambridge has achieved a significant AI milestone, creating the world’s first Large Language Model (LLM) trained completely from scratch using Federated …
1️⃣ Generative AI and Ethical Innovation: The rise of generative AI presents vast societal opportunities, but also introduces new dimensions of risk management. The Presidio AI …
AI's potential for businesses is undeniable, but the question of liability for AI-caused damages is a complex issue. Generally, one is liable for intentional and negligent …
As the latest addition to the Microsoft 365 product family, the AI-based Copilot is likely to attract scrutiny from data protection authorities. However, businesses should not let …
How can the EU ensure a level playing field and avoid undue influence from Big Tech? How can European start-ups balance their need for scaling up with their commitment to open and …
As a professional or researcher, you may often need to find and use information from various sources online. However, not all sources and information are reliable and accurate …
Today is 12th September 2025, the day the Data Act comes into force across the EU. As we step into this new era, we might find ourselves reminded of May 2018, when the GDPR …
The rise of artificial intelligence has opened doors to numerous advancements, one such being deepfakes. These AI-generated audio, video, or other digital content have brought …
In a recent post, I explored the risks associated with deepfakes. Today, I'd like to flip the coin and delve into the potential benefits and opportunities that this technology …
Deepfakes are hyper-realistic media content, be it photos, videos, audio or even texts, deliberately created or manipulated with AI. Deepfakes deploy a variety of techniques from …
In this article, we navigate through the critical aspects of data protection during the deployment of AI and Large Language Models (LLMs), highlighting real-world challenges and …
The potential of AGI to replace parliamentary decision-making has been a topic of discussion in recent research. Ponce (2017) suggests that the European Parliament's …
The Member States of the EU unanimously approved the Regulation laying down harmonized rules on Artificial Intelligence (AI Regulation, AI Act) in the Committee of Permanent …
The General Data Protection Regulation (GDPR) 🇪🇺 has set stringent protections for certain categories of personal data considered sensitive. These special categories of data are …
The Bavarian State Office for Data Protection shared a checklist for the development and deployment of Artificial Intelligence applications. It is only available in German and not …
The Global Risks Report 2024 by the World Economic Forum offers a comprehensive analysis of global risks, capturing insights from 1,500 experts worldwide. 🔍 Key Findings: 1 …
For those of you who read scientific papers from time to time, here is some of the latest research I would like to recommend concerning privacy enhancing tools in AI. A range of …
The focus on AI isn't just about its capabilities anymore; it's about how we, as humans, can control it. 🤔 There's a delicate balance to strike. On one hand, overly strict rules …
In the attached report, Europol focuses on the examination of ChatGPT in criminal and law enforcement contexts, providing a deeper understanding of how it operates and the …
The Data Act will officially come into effect on January 11, 2024, ushering in a new era of data management. The main provisions of the Data Act will be fully applicable starting …
In November 2022, the German Data Protection Conference's "Microsoft Online Services" working group highlighted concerns about Microsoft's Data Protection Addendum for Microsoft …
On 09.03.2023, the German Federal Office for Information Security published the attached guideline. It serves as an introductory resource for developers, shedding light on …
The EU's Data Act is set to revolutionize the landscape, ensuring fair treatment and appropriate use of data. As we gear up for the implementation of this legal framework, let's …
The Conference of Data Protection Supervisory Authorities for Federal and State Governments in Germany (Datenschutzkonferenz/DSK) has issued a press release on November 29, 2023 …
In a world where AI-related legislation is on the rise, effective policymaking demands evidence, foresight, and international collaboration. AIM steps up to the challenge by …
The EU Council just greenlit the Data Act, a game-changer for fair data access. For example, car owners could choose in the future to share certain vehicle data with a mechanic or …
On July 17, 2020, the High-Level Expert Group on Artificial Intelligence (AI HLEG) concluded the development of the final ALTAI. This followed an extensive piloting process …
Following recent challenges in the AI Act Trilogue negotiations, the EU Commission has shared a compromise proposal on Sunday addressing disagreements on the regulation of …
As MEPs shape the final provisions of the landmark AI Act, recent indications suggest that the Commission has proposed new governance structures. While the specifics of this …
For those who are interested, here are some valuable resources to dive deeper into the topic and navigate the crucial tasks, especially for small enterprises: 1️⃣ Data Protection …
The EU has introduced new model contractual AI clauses, a significant step towards fostering responsible and trustworthy AI development. Starting my legal journey with …
Today, I had the privilege of presenting at IPAI, the Innovation Park Artificial Intelligence in Heilbronn. It's been a while since my last visit in 2010, and the positive …
AI models in medicine are heralded as game-changers, offering the potential for groundbreaking advancements in clinical decision-making. However, the debate continues on whether …
Recent debates have raised concerns about GDPR principles hindering AI research and applications. The CNIL's (French data protection authority) response? AI can coexist with GDPR …
Here's a breakdown: 📋 How to Use Them? Public organizations considering AI procurement can voluntarily adopt these clauses. Their use is based on a case-by-case assessment of …
As of June 2023, the EU AI Act has entered the trilogue stage, a crucial step in the legislative process. During this phase, representatives from the European Commission, the …
In the age of digital transformation, the power of AI is undeniable. However, it's essential to address a growing concern: the risk of unauthorized reidentification. This risk is …
🧐 Key Insights The EU has created frameworks for risk management, with two key tools: GDPR's Data Protection Impact Assessment (DPIA) in 2016 and the EU Parliament's proposal of …
Yes, large language models (LLMs) hallucinate, a concept popularized by Google AI researchers in 2018. Hallucinations occur when AI-generated text seems plausible but is incorrect …
As generative AI continues to evolve, safeguarding authenticity becomes paramount. Misinformation and deception can thrive when the lines blur. By establishing mechanisms to …
While the Data Governance Act creates the processes and structures to facilitate data sharing by companies, individuals and the public sector, the Data Act clarifies who can …
The WEF's Chief Risk Officers Outlook arrives amid economic instability and geopolitical tensions. Chief risk officers evaluate global volatility across five dimensions …
🔐 Zoom takes a proactive step by updating its terms to ensure customer calls won't be used for AI training without consent. https://lnkd.in/eefSZidm 📖 NYU's Carter Journalism …
Check out this opinion paper from Confederation of European Data Protection Organisations (CEDPO) on the potential impact of the EU's Artificial Intelligence Act (AI Act) on the …
However, when it comes to training your in-house AI tools and utilizing your existing (personal) data pool, it's crucial to pay close attention to Art. 6 (4) GDPR. Let's outline …
The linked study of the Panel for the Future of Science and Technology (STOA, European Parliament) presents the current state of collaborative robotics, its benefits, and its …
The European Parliament adopted its AI Act negotiating position with 499 votes in favor, 28 against, and 93 abstentions. The rules aim to ensure EU-aligned AI development …
Key Points: - The AI Act aligns with the future OECD definition of 'Artificial intelligence system' (AI system). - Extensive list of banned systems, including biometric …