We asked the same service question to the chatbots of ten major Dutch organisations. Three explicitly tell you that you are talking to AI (NS, Ziggo and CZ), two call themselves a chatbot without using the word AI, three choose softer labels such as digital assistant or digital service employee, and two had no publicly reachable chat. From 2 August 2026, Article 50(1) of the EU AI Act requires that people know they are interacting with an AI system.
Deploying Microsoft 365 Copilot makes you a deployer under the EU AI Act, and since 2 February 2025 you must ensure a sufficient level of AI literacy for everyone working with it. The complete guide: which roles need training, how to record evidence per role, and how Copilot fits into your broader AI governance.
The Dutch Data Protection Authority coordinates algorithm supervision, the RDI covers the technical side, sectoral regulators keep their own turf and Brussels supervises GPAI itself. What is already settled, what still depends on the Dutch implementation act, and who will actually send the letter.
Yes. AI agents are AI systems under Article 3(1) of the AI Act, autonomy is literally part of the definition. There is no separate agent regime. What this means per risk category, who carries what when agents run on GPAI models, and what changes when an agent acts autonomously.
Fine-tune a model, white-label a purchased tool or repurpose an AI system, and under Article 25 of the AI Act you may become the provider yourself. The three routes explained, with the practical reality of GPT fine-tunes, rebranded tools and RAG applications.
From 2 August 2026, providers of AI systems that generate synthetic audio, image, video or text must ensure the output is marked in a machine-readable format and detectable as artificially generated or manipulated. The duty sits with the provider, the solution must be effective and interoperable, and systems already on the market have a backstop until 2 December 2026.
From 2 August 2026, anyone who uses AI to generate or manipulate text that is published to inform the public on matters of public interest must disclose that the text is artificial. The duty sits with the deployer, the scope is narrow, and it falls away when a human holds editorial responsibility after substantive review. Text is not a deepfake.
The transparency obligations under Article 50 are enforceable from 2 August 2026 and were not postponed by the Digital Omnibus. A breach can lead to a fine of up to 15 million euro or 3 percent of total worldwide annual turnover, whichever is higher. Enforcement sits with the national market surveillance authorities. The duty is direct: disclose the use of AI clearly and in time, and build an evidence layer.
From 2 August 2026, anyone who provides an AI system that interacts directly with people must design that system so the person knows they are dealing with AI. The duty sits with the provider, it is a design duty, and the disclosure must be clear and distinguishable at the latest at first interaction. Disclosing alone is not enough if the chatbot steers users.
From 2 August 2026, anyone who uses AI to generate or manipulate a deepfake must disclose that the content is artificially generated. This covers image, audio and video that appear authentic, not text. The duty sits with the deployer, the disclosure must be clear and given at the latest at first exposure, and a lighter form applies to artistic or satirical work.
Article 50(3) of the AI Act requires deployers to inform people when a system for emotion recognition or biometric categorisation is applied to them. But transparency is not the first question: many of these uses are already prohibited under Article 5. So test the prohibition first, and only then the disclosure duty that applies from 2 August 2026.
The Digital Omnibus postpones the high-risk obligations for standalone Annex III systems to 2 December 2027 and softens the wording of Article 4 AI literacy, but leaves Article 50 transparency in place from 2 August 2026. The European Parliament endorsed the text on 16 June 2026 and the Council gave it final green light on 29 June 2026, but as of mid-July 2026 it is not yet in the Official Journal (publication is expected before 30 July). Until publication, the original AI Act remains the binding law. Plan against the original dates and use any postponement to get ahead, not to delay.
While the Digital Omnibus pushed high-risk AI deadlines into 2027 and 2028, the Article 50 transparency obligations were left untouched and still apply from 2 August 2026.
The Digital Omnibus pushes most high-risk AI Act obligations to December 2027, but transparency rules and the AI literacy duty stay on their original timeline.
The Commission's draft guidelines set out two routes to high-risk status under Article 6: products and safety components under Annex I, and standalone systems in the sensitive domains of Annex III.
LinkedIn Recruiter and Talent Insights deploy AI heavily: Recommended Matches, AI-assisted messaging, Hiring Assistant. Almost every EU employer uses it — almost none have it in their AI register.
The European Commission's draft guidelines make HR AI more concrete: CV ranking, targeted job ads, assessments and worker management now need defensible classification.
HiBob's Bob is popular with Dutch scale-ups thanks to UX and flexibility. The AI layer (Bob AI, talent scoring, sentiment) changes the classification question under the EU AI Act.
AFAS deliberately takes a cautious position on AI. For Dutch employers using Profit and Insite that is reassuring on the surface — but the classification question does not disappear.
A structured overview of the eight domains of Annex III AI Act based on the Commission guidelines of 19 May 2026, with use cases, examples of high-risk systems and the application of the Article 6(3) filter per domain.
On 19 May 2026 the European Commission released its draft guidelines for the classification of high-risk AI systems. A deep dive into the Article 6(3) filter, the eight Annex III domains, and the three pitfalls every provider and deployer needs to understand.
Domain 1 of Annex III AI Act covers three use cases: remote biometric identification, biometric categorisation and emotion recognition. The Commission guidelines draw sharp distinctions between what is prohibited under Article 5, what is high-risk and what is out of scope.
Domain 2 of Annex III AI Act covers six use cases where AI functions as a safety component in essential services. The Commission guidelines connect this closely to the NIS2 and CER directives.
Domain 3 of Annex III AI Act covers four use cases touching the entire education process. The Commission guidelines sharply distinguish what falls within high-risk and what counts as pedagogical support.
Deep-dive on domain 4 of Annex III AI Act: recruitment, selection and worker management. Which HR-tech is high-risk, what falls within the Article 6(3) filter, and what this means for recruitment, performance management and task allocation.
Domain 5 of Annex III AI Act covers four diverse use cases: public benefits, creditworthiness, life and health insurance, and emergency call triage. The Commission guidelines provide sharp delineation per use case, with particular attention to banks and insurers under CRR and Solvency II.
Domain 8 of Annex III AI Act covers two very different use cases: AI supporting judges, and AI intended to influence elections or referendums. The Commission guidelines set sharp boundaries.
Domain 6 of Annex III AI Act has five use cases for AI in investigation and policing. The Commission guidelines sharply delineate what is prohibited, what is high-risk and what is outside scope.
Domain 7 of Annex III AI Act covers four use cases touching the entire chain of admission to Europe. The Commission guidelines connect this closely to Schengen, EES, ETIAS and EURODAC.
A practical overview of AI Act deadlines for 2026, 2027 and 2028, separating what already applies, what the current law says and what may shift under the Digital Omnibus agreement.
The Council and European Parliament reached a provisional political agreement on the Digital Omnibus AI Act on 7 May 2026. These are the new deadlines, the open Article 4 question and the actions organisations should take now.
Personio is dominant in DACH and growing in the Dutch SME market. The AI roadmap (Personio Conversations, AI sourcing, automation) changes the classification question. Practical analysis.
SAP has built an enterprise AI stack with Joule and the Talent Intelligence Hub that runs through HR processes end to end. Practical classification and vendor questions under Annex III point 4.
HireVue scores candidates on responses, language and game-based assessments. Under the EU AI Act that means: defensive starting point is high-risk. Practical analysis and vendor questions.
On 20 April 2026 the Dutch State Secretary Aerdts launched the public consultation on the Implementation Act of the AI Regulation. The law defines who supervises the EU AI Act in the Netherlands — and why organisations should pay attention now.
Recruitee positions itself as a user-friendly ATS, but Smart Capture, candidate matching and Hire AI hit Annex III point 4(a) directly. Practical analysis and vendor questions.
Article 9 requires a continuous risk management system for high-risk AI. Here is what providers must document, test, mitigate, and review before go-live.
Bullhorn is dominant in Dutch recruitment agencies and staffing firms. With AI Recruiter, automation and the GPT layer almost every Bullhorn deployment is within Annex III point 4(a) — for the agency and their clients.
Article 50 splits transparency duties between providers and deployers. Here is who must label AI content, disclose deepfakes, and inform users under the EU AI Act.
What Article 10 demands of high-risk AI providers, paragraph by paragraph: the eight data governance areas, quality thresholds, bias rules, and what to document.
Visma is strongly represented in the Dutch SME and (semi-)public sector. The HR portfolio (Visma Recruit, Talent, Verzuim, Loon) is more conservative on AI — but the classification question does not disappear.
Article 14 requires human oversight for all high-risk AI systems. Here is what it means in practice: who oversees, what authority they need, and how to avoid automation bias.
Historical overview of the European Parliament position of 26 March 2026 on the AI Act Omnibus. A provisional political agreement was reached on 7 May 2026, so read this article as context for the earlier Parliament line.
On 29 July 2025 the European Commission published guidelines on when software qualifies as an AI system under the AI Act. Seven elements determine whether your tool falls under the regulation.
On 19 March 2026, the IMCO and LIBE committees of the European Parliament voted in favour of a package of amendments to the AI Act. High-risk AI systems get more time, but obligations already in force remain fully applicable.
Lever positions itself as Talent Relationship Management — CRM approach to recruitment. AI features in sourcing, nurture and analytics hit Annex III point 4(a). Practical analysis.
Greenhouse is known for structured interviewing and bias mitigation. The AI layer (matching scores, AI writing, sourcing recommendations) makes the classification question relevant nonetheless.
BambooHR is popular with SMEs and scale-ups for its simplicity. The AI layer rolled out since 2024 (Ask BambooHR, AI screening, performance) changes the compliance question under the EU AI Act.
European Commission missed its February 2026 deadline for high-risk AI guidance. What this means for your compliance timeline and what next steps to take.
Analysis of the original Digital Omnibus on AI proposal. A provisional political agreement was reached on 7 May 2026, so read this article as background to the earlier Commission position.
From the European AI Office to national supervisors: discover which agencies and authorities will enforce the EU AI Act and what this means for your...
What the EBA letter means for banks: how AI Act duties for credit scoring overlap with CRR/CRD and DORA, and where you must add controls instead of duplicating them.
How do governments conduct a FRIA (Fundamental Rights Impact Assessment)? Practical case study of a rights assessment for high-risk AI in the public...
How government organizations can inventory and classify their AI systems according to the EU AI Act, including a practical approach to risk classification.
A complete handbook for government organizations on EU AI act implementation, focusing on deadlines, obligations, and practical compliance steps for the...
Eleven months after the EU AI Act entered into force: the first prohibitions apply, employees must be AI-literate, and Brussels is flooded with consultations. Here is what happened and what it means for your organization.
Eight sectors classified as high-risk under the EU AI Act: biometrics, critical infrastructure, HR, and more, with practical compliance tips per sector.
An in-depth analysis of the EU AI act implementation in 2025, focusing on initial applications, Commission guidelines, and implications for businesses.
Discover how the EU AI Act affects start-ups and SMEs. From compliance requirements to practical tips, including exceptions and supporting measures for...
