EU AI Act for Moroccan companies: 9 checks before serving an EU client

Nine checks to determine the EU AI Act scope, your role, risk level, applicable obligations, and evidence to prepare before serving an EU client.

EU AI Act for Moroccan companies: 9 checks before serving an EU client

The EU AI Act can apply to Moroccan companies even when they have no establishment in the European Union. The regulation covers providers placing an AI system or a general-purpose AI model on the EU market, and it can cover providers or deployers located in a third country when the output produced by the system is used in the Union. Using an AI tool in Morocco alone does not mean that every EU obligation applies.

A practical assessment should follow nine checks: territorial trigger, legal role, object supplied, exclusions, risk level, applicable timeline, transparency, contractual chain, and operational evidence. This framework is not legal advice. It prepares the facts, documents, and questions a Moroccan company should clarify with its EU client and qualified counsel.

What is the EU AI Act and why use the consolidated text?

The EU AI Act is Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence. Duties vary by actor, type of system, use, and risk. The original regulation has been amended. As of 19 November 2026, the reference source is the consolidated text on EUR-Lex, together with current Commission guidance.

The distinction matters for dates. Prohibitions, definitions, and AI literacy provisions have applied since 2 February 2025. Governance rules and obligations for general-purpose AI models have applied since 2 August 2025. Many provisions and Article 50 transparency rules have applied since 2 August 2026. Following the 2026 amendment, selected high-risk rules apply from 2 December 2027, while high-risk AI embedded in certain regulated products follows 2 August 2028. Recheck the Commission’s official AI Act timeline FAQ before relying on a date.

Check 1 — What connects the activity to the European Union?

Article 2 covers providers placing an AI system on the EU market or putting it into service, regardless of where the provider is established. It also covers providers and deployers in a third country when system output is used in the Union. Importers, distributors, product manufacturers, and authorised representatives are included in their respective situations.

Map the actual delivery path: contracting entity, location of users, name under which the system is supplied, party deciding its purpose, place where output is used, and whether the service becomes part of the client’s product. Incidental, unforeseeable, or unauthorised downstream use should not automatically be treated like an organised placing on the market; the facts and contracts matter.

  • Is the software offered directly to users in the Union?
  • Does the EU client receive a human service or an AI system it can operate under your name?
  • Are analyses, scores, content, or decisions produced in Morocco used in an EU process?
  • Does the client change the intended purpose or market the system under its own trademark?

Check 2 — Are you a provider, deployer, importer, or contractor?

The role is not defined by a sales label. A provider develops an AI system, or has one developed, and places it on the market or puts it into service under its name or trademark. A deployer uses a system under its professional authority. Importers and distributors have different supply-chain duties. An integrator can become a provider when it applies its name, makes a substantial modification, or changes the intended purpose under the conditions set by the regulation.

Create a matrix for every use case: entity, role, system, model, intended purpose, users, territory, and responsibility. The same Moroccan group may deploy a purchased assistant internally and provide a different system to a French client. Qualification is case-specific.

Check 3 — Are you supplying an AI system, a GPAI model, or an AI-assisted human service?

The AI Act distinguishes an AI system from a general-purpose AI model. A company that uses a model API to build a business assistant may provide a system without being the provider of the underlying GPAI model. Duties for the model provider and those for the system provider or deployer are not interchangeable.

Describe the object without marketing language: inputs, outputs, autonomy, inference logic, human decisions, interfaces, model version, and intended conditions of use. Record what the customer receives: application access, API, model, code, report, recommendation, or a human service. This prevents assigning GPAI duties to a simple integrator or overlooking that a product is supplied under your own brand.

Check 4 — Does an exclusion or exception actually apply?

The regulation includes exclusions for certain activities carried out exclusively for military, defence, or national-security purposes, for systems and models specifically developed and put into service solely for scientific research, and for natural persons using AI in a purely personal non-professional activity. Research, testing, and development before placing a system on the market also receive specific treatment subject to the regulation’s conditions.

Calling a production service a “pilot” does not create an exclusion. If the system already influences a client service or real users, verify whether it remains limited to research or testing. Document its duration, users, data, environment, and permitted decisions.

Check 5 — Is the use prohibited, high-risk, or subject to transparency?

Start with prohibited practices. The regulation bans defined forms of manipulation, exploitation of vulnerabilities, social scoring, and other uses under precise conditions. Do not classify a use only by industry. Assess the intended purpose, affected people, and actual effect.

High-risk systems are identified through Article 6, Annex I for certain regulated products, and Annex III for specified uses. Depending on the exact conditions, areas include employment, education, access to essential services, biometrics, critical infrastructure, law enforcement, migration, and justice. A narrow procedural or preparatory task may fall within a nuance set by the regulation, so classification requires a documented analysis rather than a keyword.

Interactive and generative systems can also trigger Article 50 transparency obligations. The Commission’s official transparency guidance addresses information for people interacting directly with AI and machine-readable marking of certain synthetic content.

Check 6 — Which duty applies on which date?

Do not build a compliance plan from an outdated article. For every selected duty, record the article, relevant role, type of system, application date, and official source consulted. Prohibitions, AI literacy, GPAI, transparency, and high-risk systems do not follow the same timetable.

As of 19 November 2026, Article 50 transparency rules and several enforcement powers apply. Selected high-risk provisions follow the new 2027 and 2028 dates. A future standard, guideline, or amendment can change how compliance is demonstrated. Assign an owner to regulatory monitoring and retain the consultation date for every source.

Check 7 — What technical and human measures can you prove?

A broad policy is not evidence by itself. Link every applicable requirement to an artefact: inventory, intended purpose, data, performance criteria, tests, limitations, human oversight, logs, incident handling, cybersecurity, model versions, and user information. Exact duties vary by classification; do not copy the entire high-risk chapter onto a system outside that category.

Three Kanteek resources support this work:

Add meaningful human review where an error can affect a right, access, contractual duty, or important decision. The reviewer needs time, information, and authority to change or stop the output. A theoretical approval button is insufficient.

Check 8 — Is an EU representative required and what should contracts say?

An EU authorised representative is not universally required for every Moroccan company using AI. Article 22 requires one for third-country providers before making certain high-risk systems available on the Union market. Article 54 sets a requirement for certain third-country providers of GPAI models, with a defined exception for some free and open-source models that do not present systemic risk.

The contract chain should reflect the real roles. Identify the model provider, host, integrator, customer, possible importer, and owner of updates. State who prepares documentation, supplies information, manages versions, reports incidents, responds to authority requests, and decides whether a change is substantial.

A clause stating that each party “complies with all applicable law” is not a deliverable. Specify documents, deadlines, contacts, reasonable audit rights, and change rules. A contract cannot turn missing evidence into available evidence.

Check 9 — What file should exist before serving the customer?

Prepare a file proportionate to the system. It may include:

  • a territorial-scope sheet covering customers, users, and where output is used;
  • a role matrix for each entity in the chain;
  • a system description, underlying model, intended purpose, and limitations;
  • a classification note with articles, annexes, assumptions, and legal validation;
  • an obligation calendar with sources and consultation dates;
  • transparency messages and tests in every required language;
  • evaluation reports, version logs, and human-oversight criteria;
  • data, provider, and access maps;
  • contracts, contacts, incident processes, and reversibility evidence.

The file must follow the version actually supplied. A change of model, purpose, population, or integration may alter the role, risk, or evidence. Include compliance review in the change process, not only before the first contract.

A four-workshop method

Workshop 1 — Scope and roles

Bring together business, product, engineering, and legal representatives. Map the system, output, and user journey. Produce the role matrix and list of assumptions to confirm with the customer.

Workshop 2 — Classification and timeline

Check exclusions, prohibited practices, high risk, transparency, and GPAI. Tie every decision to the consolidated text and a current application date.

Workshop 3 — Controls and evidence

Turn applicable duties into product controls, tests, user information, human oversight, logging, and deliverables. Do not add a control without an owner and success criterion.

Workshop 4 — Contract and change

Align duties with contracts, suppliers, and maintenance. Define who informs whom when a model version, incident, intended purpose, or authority request changes.

Common mistakes

  • assuming that no EU establishment means the AI Act cannot apply;
  • assuming the opposite, that every use of AI in Morocco falls under the regulation;
  • confusing the GPAI model provider, system provider, and deployer;
  • using a timeline published before the 2026 amendment;
  • classifying high risk solely from an industry label;
  • appointing an EU representative before establishing the role and requirement;
  • adding an “AI-generated” label without checking the full scope of Article 50;
  • keeping a policy but no version-linked evidence;
  • leaving contracts silent on models, updates, and incidents.

FAQ — EU AI Act for Moroccan companies

Is a Moroccan agency automatically covered when it uses ChatGPT for drafting?

No. Assess the service supplied, place where output is used, the agency’s role, and the integration. Internal use in Morocco does not automatically trigger every duty. Offering a system under the agency’s own brand to EU users can create a different connection.

Is the EU customer always solely responsible?

No. Responsibility depends on roles. The customer may be a deployer, importer, or become a provider after certain changes. The Moroccan company may remain the system provider. Contracts should allocate tasks without contradicting statutory responsibilities.

Do all high-risk requirements already apply?

No. The timetable is progressive and changed in 2026. Some rules already apply, while selected high-risk provisions follow dates in 2027 and 2028 depending on category. Check the current official FAQ and consolidated text for the precise case.

Must a chatbot say that it is AI?

Within its scope, Article 50 requires people to be informed when they interact directly with an AI system unless this is obvious to a reasonably well-informed and attentive person. Test the wording, timing, and accessibility in the actual interface.

Must every Moroccan company appoint an EU representative?

No. The requirement depends on being a third-country provider and the object supplied. Articles 22 and 54 cover certain high-risk systems and certain GPAI models respectively. Confirm qualification before signing a mandate.

Prepare a decision, not an improvised certification

The EU AI Act for Moroccan companies is not a universal checklist. The task is to establish the scope, role, classification, timeline, and controls corresponding to the system actually supplied to EU customers or users.

Kanteek supports use-case framing, evaluation, governance, and technical integration through its Consulting & Strategy and Artificial Intelligence services. Final legal analysis belongs with qualified counsel; technical work makes the facts, versions, and evidence available for that decision.