Skip to main content
Praxikon

Commission guidelines under Article 96 AI Act: what exists, what is missing and what they are worth

··16 min read

The legal adviser of a municipality receives three questions. Is the calculation tool for parking permits an AI system? Must the chatbot on the website disclose that it is AI? Is the model that flags applications for social assistance a high-risk AI system? For each question the European Commission has guidelines, but two have been adopted and the third is a draft. May the municipality build on them?

Below, Article 96 paragraph by paragraph; the examples are fictional.

In short

Article 96(1) requires the Commission to develop guidelines on the practical implementation of the Regulation, on seven topics. Regulation (EU) 2026/1744 added the obligations of deployers and a topic with a deadline of 1 August 2027 (point (g)). The guidelines on prohibited practices, the definition of an AI system, general-purpose AI models and Article 50 have been adopted; those on high-risk classification and serious incidents are drafts.

Guidelines are not binding; authoritative interpretation rests with the Court of Justice. In our reading they are nonetheless the best indication of how a supervisory authority will apply the rules.

The content of the guidelines themselves is covered elsewhere: the definition of an AI system, social scoring, emotion recognition in the workplace, the Article 50 guidelines, the missed deadline for high-risk classification, serious incidents and the obligations of model providers. Binding additions are covered in the post on Articles 97 and 98.

Paragraph 1: seven topics

The Commission "shall develop guidelines on the practical implementation of this Regulation", in particular on:

  • point (a) the application of the requirements and obligations referred to in Articles 8 to 15 and in Articles 25 and 26: requirements for high-risk systems, the AI value chain and deployers;
  • point (b) the prohibited practices referred to in Article 5;
  • point (c) the practical implementation of the provisions related to substantial modification;
  • point (d) the practical implementation of the transparency obligations laid down in Article 50;
  • point (e) the relationship with the Union harmonisation legislation listed in Annex I and with other relevant Union law, including consistency in their enforcement;
  • point (f) the application of the definition of an AI system in Article 3, point (1);
  • point (g) Article 8(2), Article 9(10) and Article 17(3), to avoid duplication with the legislation listed in Section A of Annex I; to be published by 1 August 2027.

The Digital Omnibus amended paragraph 1 on three points (Article 1, point 36): Article 26 was added to point (a), point (g) is new, and the second subparagraph was replaced. Point (g) concerns integrating AI-specific risks into existing risk and quality management systems (recital 37 of Regulation (EU) 2026/1744). Article 96 sits in Chapter X and has applied since 2 August 2026 (Article 113, second paragraph). The first guidelines appeared in 2025, with Article 96(1) as their basis; in our reading nothing in the Regulation prevents the Commission from giving its interpretation earlier, precisely because that interpretation is not binding. That fits the Regulation itself: Article 6(5) and Article 73(7) require guidelines before 2 August 2026.

Second and third subparagraphs: with the Board, and with an eye on local authorities

The Commission shall involve the Board and pay particular attention to the needs of SMEs, including start-ups, and SMCs, of local public authorities and of the sectors most likely to be affected by the Regulation. The Board and the SMCs were added by the Omnibus; local public authorities were there from the start.

The third subparagraph requires due account of the generally acknowledged state of the art and of the standards and common specifications referred to in Article 40 and Article 41 or under Union harmonisation law.

Paragraph 2: updates

At the request of the Member States or the AI Office, or on its own initiative, the Commission shall update guidelines when deemed necessary. In our reading a practical problem is therefore best raised through a consultation or through the national government.

Other articles that mandate guidelines

Article 96 is the general mandate. Four provisions ask for guidelines themselves:

  • Article 6(5): high-risk classification, with practical examples and "in line with Article 96", no later than 2 February 2026 (see also recital 53);
  • Article 63(1): the elements of the quality management system with which SMEs may comply in a simplified manner, without a deadline (see recital 146, which still referred to microenterprises);
  • Article 72(3): guidance, including a template, on the post-market monitoring plan, by 2 September 2027;
  • Article 73(7): dedicated guidance on reporting serious incidents, to be issued by 2 August 2025.

In our reading, not guidelines in this sense are the template for the fundamental rights impact assessment under Article 27(5) (although the Commission labels the package with its draft explanatory notice "guidelines with a template"), the AI literacy examples under Article 4(2), codes of practice (Article 56 and Article 50(7)) and implementing acts (adopted under the procedure of Article 98).

What exists on 22 November 2026

TopicLegal basisDate and numberStatus
Prohibited AI practicesArticle 96(1)(b)Content approved 4 February 2025 (C(2025) 884 final); adopted 29 July 2025, C(2025) 5052 finalAdopted
Definition of an AI systemArticle 96(1)(f)Content approved 6 February 2025 (C(2025) 924 final); adopted 29 July 2025, C(2025) 5053 finalAdopted
Obligations of providers of general-purpose AI modelsArticle 96(1)Content approved 18 July 2025 (C(2025) 5045 final); adopted 19 November 2025, C(2025) 7719 finalAdopted
Transparency obligations of Article 50Article 96(1)(d)Content approved 20 July 2026 (C(2026) 5054 final); adopted 5 November 2026, C(2026) 7702 final; Official Journal 12 November 2026, C/2026/6005Adopted
High-risk classificationArticle 6(5), with Article 96Draft 19 May 2026; consultation closed 23 July 2026; 2 February 2026 deadline missedDraft; final by the end of 2026 according to the Commission
Reporting serious incidents, with reporting templateArticle 73(7)Draft 26 September 2025; consultation closed 7 November 2025; 2 August 2025 deadline missedDraft
Template for the fundamental rights impact assessmentArticle 27(5) (in our reading not an Article 96 guideline)Draft with draft explanatory notice 16 November 2026; feedback until 15 January 2027Draft

Cite a guideline by C-number and date: in the guidelines on general-purpose AI models, the adopted text is numbered differently from the July 2025 text.

What is still missing

On its overview the Commission lists as work in preparation: guidelines on the obligations of providers and deployers of high-risk AI systems (point (a)), responsibilities along the AI value chain (point (a), Article 25), substantial modification (point (c)), a voluntary template for post-market monitoring (Article 72), the simplified quality management system for SMEs and SMCs (Article 63), and the interplay with other EU legislation (point (e)), including joint guidelines of the Commission and the European Data Protection Board on the AI Act and EU data protection law. The FRIA template is listed there as well, as guidelines with a template; that draft has been out since 16 November 2026. The Commission gives no dates. For point (g) we found nothing yet; the deadline is 1 August 2027. The overview lists the serious incident reporting template as published, but the link does not work and we found no adopted version; we rely on the draft of 26 September 2025.

For medical devices there is a question-and-answer document of the AI Board and the Medical Device Coordination Group (AIB 2025-1 / MDCG 2025-6, June 2025). It states that it is not a European Commission document and does not reflect the Commission's official position, although the overview calls it published guidance; in our reading it is not a guideline under point (e).

What guidelines are in law, and what they are not

Not binding. All adopted guidelines say so themselves: they are not binding and any authoritative interpretation of the AI Act may ultimately only be given by the Court of Justice (prohibited practices paragraph 5, definition paragraph 7, Article 50 paragraph 5, general-purpose AI models paragraph 9). A national court can ask the Court for an interpretation (Article 267 TFEU), and a judgment can overtake a guideline.

Guiding for enforcement. The Commission writes that the guidelines should assist competent authorities in their enforcement activities (prohibited practices paragraph 4) and provide practical guidance to competent authorities (Article 50 paragraph 3), that it will base its enforcement action on the guidelines for general-purpose AI models (paragraph 9), and, of the draft on high-risk classification, that it reflects the Commission's interpretation and will guide enforcement, although the consultation is still to be processed (Commission page). The Regulation also gives them a role: a market surveillance authority that doubts a provider's reliance on Article 6(3) evaluates the system "based on the conditions set out in Article 6(3) and the Commission guidelines" (Article 80(1)), and Member States lay down penalty rules taking into account the guidelines issued pursuant to Article 96 (Article 99(1)).

What this means. In our reading, an organisation that demonstrably follows an adopted guideline stands strong before a supervisory authority, but it is not a free pass: a court may decide otherwise, and a draft has not been adopted yet and may change after the consultation. Record which version you used, with C-number, date and paragraph.

What the municipality gains

The calculation tool for parking permits is assessed against the adopted guidelines on the definition. Systems based on the rules defined solely by natural persons to automatically execute operations are not covered (recital 12, quoted in paragraph 40 of the guidelines). The chatbot falls under Article 50(1), which has applied since 2 August 2026. That obligation rests on the provider. If the municipality buys an off-the-shelf chatbot that the supplier offers under its own name, the supplier is the provider. If the municipality builds it itself or has it developed, and puts it into service under its own name, it is provider and deployer at the same time (Article 3, points (3) and (4)). The adopted Article 50 guidelines are the reference.

The flagging model for social assistance is in principle high-risk (Annex III, point 5(a)), unless Article 6(3) provides an exception, which never applies where the system performs profiling. The guidelines on this are a draft; in our reading the municipality uses the draft as the best available interpretation and records its own reasoning. For its obligations as deployer (Article 26), which apply from 2 December 2027, there are no guidelines yet. The draft template under Article 27(5) helps with the fundamental rights impact assessment it must carry out as a body governed by public law; for a system already placed on the market or put into service before 2 December 2027, the high-risk obligations apply under Article 111(2) only after a significant change in its design, and for public authorities by 2 August 2030 at the latest. As a public authority it must also register its use in the EU database (Article 26(8) and Article 49(3)). If it substantially modifies the supplier's system, it may become a provider (Article 25); precisely those guidelines are still missing.

Bank or insurer

For a bank's chatbot, the adopted Article 50 guidelines apply. Its credit scoring of natural persons is in principle high-risk (Annex III, point 5(b), with an exception for detecting financial fraud), as is risk assessment and pricing in life and health insurance (point 5(c)); for those there is the draft. As deployers of such systems they too must assess the impact on fundamental rights (Article 27(1)), for which the draft template helps. If a bank builds its own system on another party's model, the guidelines on general-purpose AI models help determine its role. For the interplay with financial services law (point (e)) there is no Commission guideline yet. For high-risk AI systems placed on the market, put into service or used by financial institutions, the market surveillance authority is the financial supervisory authority, in so far as this is in direct connection with the provision of their financial services (Article 74(6)); in our reading it will use the guidelines like any other market surveillance authority. For the chatbot this does not follow from paragraph 6. The Dutch bill for the Uitvoeringswet AI-verordening is still a draft.

Where the GDPR comes in

For the GDPR the European Data Protection Board issues its own guidelines (Article 70(1)(e) GDPR), a different category from a different author. According to the Commission, both are working on joint guidelines on the interplay. Until those exist, the municipality assesses the processing of personal data in the flagging model separately, in a data protection impact assessment where one is required (Article 35 GDPR).

To see which AI Act and GDPR obligations apply to your use case, try the case check.

Frequently asked questions about Article 96

Sources

EUR-Lex: Regulation (EU) 2024/1689 (AI Act), original text, Article 96 and recitals 12, 53 and 146 (13 June 2024, published 12 July 2024, accessed 22 November 2026)
EUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI), Article 1, point 36, and recital 37 (8 July 2026, published 24 July 2026, accessed 22 November 2026)
European Commission: Draft Commission guidelines on the classification of high-risk AI systems (19 May 2026, accessed 22 November 2026)
European Commission: Draft guidance and reporting template on serious AI incidents (Article 73), consultation (26 September 2025, consultation until 7 November 2025, accessed 22 November 2026)
European Commission: Consultation on the draft template for the fundamental rights impact assessment (Article 27(5)) (16 November 2026, accessed 22 November 2026)
AI Board and Medical Device Coordination Group: AIB 2025-1 / MDCG 2025-6, Interplay between the MDR and IVDR and the Artificial Intelligence Act (June 2025, accessed 22 November 2026)
EUR-Lex: Treaty on the Functioning of the European Union, Article 267 (consolidated version 2016, accessed 22 November 2026)
EUR-Lex: Regulation (EU) 2016/679 (GDPR), Articles 35 and 70 (27 April 2016, accessed 22 November 2026)
Internetconsultatie.nl: Consultation on the Dutch AI Act Implementation Act (Uitvoeringswet AI-verordening), closed (consultation 20 April to 1 June 2026, accessed 22 November 2026)

Newsletter

Every Tuesday, the AI Act week ahead in 5 minutes

A practical briefing on deadlines, new guidance and enforcement, so you know what matters this week. No spam and you can unsubscribe in one click.

Practical and short · No spam · One-click unsubscribe

Written by

Zahed Ashkara, freelance AI and privacy consultant and legal adviser

Zahed Ashkara

Freelance AI and privacy consultant and legal adviser · AI governance & safety

Zahed Ashkara is a freelance AI and privacy consultant and legal adviser for the GDPR and the EU AI Act. Organisations hire him as AI governance consultant, privacy counsel, external DPO or AI governance lead, have him carry out a DPIA, FRIA or AI Act classification, or book an in-company training. He focuses on AI governance and safety, including AI agents. He works exclusively as a freelancer, through Embed AI (trade name Praxikon). He is a Certified AI Compliance Officer (CAICO) and has been a member of the NEN standards committee Artificial Intelligence & Big Data since July 2026.

Follow on LinkedInView expert profile

Use these sources in your AI assistant too: Praxikon in Claude and other assistants

Real-World Stories

Discover how organizations apply this in practice

Related articles

Article 101 AI Act: Commission fines for providers of general-purpose AI models

Article 101 of the AI Act lets the European Commission fine providers of general-purpose AI models, up to 3% of worldwide annual turnover or 15 million euro. Which four kinds of conduct can be fined, which safeguards apply, what Implementing Regulation (EU) 2026/1755 provides, how it relates to Articles 99 and 100, and whether a municipality or bank can receive such a fine.

EU AI ActArticle 101Fines

Article 55 AI Act: general-purpose AI models with systemic risk and the requirements for external evaluators

Article 55 of the AI Act gives providers of the most capable AI models four additional obligations: model evaluation, risk assessment, serious incident reporting and cybersecurity. What each paragraph says, whether the law itself requires external evaluators, what the AI Office wrote on 21 October 2026 about their qualification and independence, and what a municipality or bank buying such a model can do with it.

EU AI ActArticle 55General-purpose AI models