Skip to content
SimpleAct Logo

Guide · Legal & general counsel

As of 6 October 2026 · after the Digital Omnibus

EU AI Act for Legal Teams and General Counsel: Roles, Classification, Contracts

For the legal team the AI Act is above all a mapping exercise: which role does the organisation have for each system, which risk class follows from the purpose, what must the vendor contract say – and how is each of these decisions reasoned so that it survives scrutiny? This guide covers the duties, the assessment steps and the tasks for the next 90 days.

  • Roles under Art. 3 and Art. 25
  • Classification with reasons
  • Contract points with AI vendors
  • Checklist with progress
app.simpleact.de

Legal Logic

AI-based application screening · result

A

Deployer

role

High

Annex III point 4

12 mo

review cycle

247

audit entries

Applicable duties with reasons

  • Art. 26 · deployer obligationsDeployer of a high-risk system under Annex III point 4Applies
  • Art. 26(7) · informing workersUsed at the workplaceApplies
  • Art. 27 · fundamental rights impact assessmentPrivate deployer, not an Annex III 5(b)/(c) caseDoes not apply
  • Art. 86 · right to explanationDecision with significant effect on applicantsApplies

Audit trail

  • Hash chain verified
  • unchanged
  • Export PDF/DOCX

Sample data · Legal Logic from Professional

In short

The legal team determines the organisation’s role for each AI system (provider, deployer, importer, distributor), reviews the classification including exceptions under Art. 6(3), settles duties and information in contracts with vendors and customers, and makes sure every decision is reasoned and traceably documented. Implementation sits with business units, IT and governance.

Basics

Where the AI Act calls for legal work

The AI Act (Regulation (EU) 2024/1689) ties its duties to two features: the operator’s role and the system’s risk. Operators are providers, product manufacturers, deployers, authorised representatives, importers and distributors (Art. 3(8)). A provider develops an AI system or has it developed and places it on the market or puts it into service under its own name or trademark (point 3); a deployer uses it under its own authority (point 4). The same organisation can hold different roles for different systems – and change role for the same system over time (Art. 25).

The risk follows from the purpose. The practices in Art. 5 are prohibited. High-risk are safety components of products under Annex I that require third-party conformity assessment (Art. 6(1)) and the use cases in Annex III (Art. 6(2)) – with an exception under Art. 6(3) where there is no significant risk and one of four conditions is met. Systems that perform profiling of natural persons are always high-risk. Transparency duties under Art. 50 apply on top, regardless of the risk class.

These classifications are legal analysis in the proper sense: they depend on interpretation, facts and guidelines that are partly still open. The Commission’s guidelines on high-risk classification were published as a draft on 19 May 2026, the consultation ran until 23 July 2026, and the final version is expected by the end of 2026. Until then: give reasons for classifications, disclose assumptions, set a review date.

Then there are the interfaces. Almost every AI system touches the GDPR, many touch employment law, some the product safety or supervisory law of a sector. And every purchased system raises contract questions: what information does the vendor owe, who reports incidents, who bears which duty when the organisation modifies the system?

The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) moved the high-risk deadlines (Annex III from 2 December 2027, Annex I from 2 August 2028), rewrote Art. 4 and added two new prohibitions from 2 December 2026. Registration when relying on the Art. 6(3) exception remains mandatory, with a simplified procedure.

operator roles under Art. 3(8) AI Act
6
operator roles under Art. 3(8) AI Act
exception conditions under Art. 6(3)
4
exception conditions under Art. 6(3)
triggers for switching into the provider role (Art. 25(1))
3
triggers for switching into the provider role (Art. 25(1))

Duties

What the EU AI Act means for legal teams

The table shows the provisions that most often land on the legal team’s desk, who they address and the legal task behind each. The addressee is the organisation in its respective role.

Scroll the table sideways →

What the EU AI Act means for legal teams
TopicBasisAddresseeApplies since / fromLegal team’s task
Determine the role per AI systemArt. 3(3)–(8) AI ActAll operatorssince 1 Aug 2024 (definitions)Provider, deployer, importer or distributor – documented per system, with reasons.
Prohibited practicesArt. 5 AI ActAll operators2 Feb 2025; new prohibitions 2 Dec 2026Assess borderline cases such as workplace emotion recognition or biometric categorisation.
Promote AI literacyArt. 4 AI Act (amended)Providers and deployers2 Feb 2025Place the scope after the Omnibus: take measures, no guaranteed level.
High-risk classificationArt. 6(1), (2), Annex I, IIIProviders (classification), deployers (plausibility)2 Dec 2027 (III) / 2 Aug 2028 (I)Check the purpose against Annex III, spot profiling, give reasons for the result.
Exception and documentationArt. 6(3), (4), Art. 49 AI ActProviders2 Dec 2027Check conditions (a)–(d), document the assessment before placing on the market, registration (simplified after the Omnibus).
Switch into the provider roleArt. 25(1), (2) AI ActDeployers, distributors, importers, third parties2 Dec 2027 (III)Check branding, substantial modification and purpose change; secure the initial provider’s cooperation.
Agreements in the supply chainArt. 25(4) AI ActProviders of high-risk AI and suppliers2 Dec 2027 (III)Written agreement on information, capabilities, technical access and assistance.
Deployer obligationsArt. 26 AI ActDeployers of high-risk AI2 Dec 2027Assign the duty catalogue internally; secure information and reporting duties by contract.
Fundamental rights impact assessmentArt. 27 AI ActPublic bodies, private providers of public services, Annex III 5(b)/(c)2 Dec 2027Check the scope; use the reference to the DPIA under para. 4.
Transparency dutiesArt. 50 AI ActProviders (1, 2), deployers (3, 4)2 Aug 2026; para. 2 legacy 2 Dec 2026Review notices and labelling, place exceptions (e.g. editorial control).
Right to explanationArt. 86 AI ActDeployers (Annex III)2 Dec 2027Define the process and response templates for explanation requests.
PenaltiesArt. 99 AI ActAll operatorssince 2 Aug 2025Place the fine framework, including for incorrect or incomplete information to authorities.

Simplified overview. Providers of high-risk AI have further duties (among others Art. 9–17, 43, 47–49). Deadlines after the Digital Omnibus, Regulation (EU) 2026/1744.

In detail

Six topics for legal teams

Each topic covers the legal position, the usual questions of interpretation and a practical recommendation. Recommendations are marked as such.

01

Determine roles – per system, not per organisation

Art. 3(3), (4), (6), (7), (8) AI Act

The question “are we provider or deployer?” has no blanket answer. An organisation can be a deployer of a purchased chat assistant, a provider of an in-house scoring model and an importer of a system from a manufacturer in a third country (point 6). Whoever merely passes a system on without being its provider or importer is a distributor (point 7). Each role has its own catalogue of duties.

One sentence per system is enough for the documentation – but it has to be right. Typical errors: an internally developed tool is filed as “deployer” although it is put into service under the organisation’s own name; a partner’s white-label product is overlooked although own branding can lead to the provider role under Art. 25(1)(a) if it is a high-risk system.

  • Record the role per system, with short reasons
  • In-house development under own name: provider
  • Systems from third countries: check the importer role
02

Give reasons for classification – and check the exception properly

Art. 5, 6(1)–(4), Annex III AI Act

A proven order of assessment: first prohibited practice (Art. 5), then safety component of an Annex I product with third-party assessment (Art. 6(1)), then Annex III use case (Art. 6(2)), then transparency duties (Art. 50). The specific purpose decides, not the product.

For Annex III cases the exception under Art. 6(3) can apply where there is no significant risk to health, safety or fundamental rights and the system performs a narrow procedural task (a), improves the result of a previously completed human activity (b), detects decision-making patterns without replacing the human assessment without proper review (c), or performs a preparatory task (d). Profiling of natural persons rules the exception out. The provider must document its assessment before placing on the market and provide it on request (para. 4); registration remains mandatory after the Omnibus.

As a deployer you do not classify in the provider’s place, but you check plausibility: does the provider’s intended purpose match your actual use? If the use differs, the classification can change – up to your own provider role under Art. 25(1)(c).

  • Order: Art. 5 → Annex I → Annex III → Art. 50
  • Exception only with a documented assessment, never with profiling
  • As deployer: check intended purpose against actual use
03

Switching into the provider role

Art. 25(1), (2) AI Act

Deployers, distributors, importers or third parties are considered providers of a high-risk system and take on the duties under Art. 16 if they put their name or trademark on a high-risk system already placed on the market, make a substantial modification so that it remains high-risk, or modify the intended purpose of an AI system so that it becomes high-risk (Art. 25(1)). The initial provider is then no longer considered the provider of that system, but must cooperate closely, provide the necessary information and give technical access – unless it has clearly specified that its system is not to be changed into a high-risk system (para. 2).

For the legal team this is a contract question. Recommendation: before modifying a purchased system, check whether the provider has excluded a change into high-risk, and secure by contract the cooperation para. 2 envisages. And anchor a review in the change process before a system is used for a new purpose.

  • Three triggers: branding, substantial modification, purpose change
  • Initial provider’s cooperation duties (para. 2)
  • Check whether the initial provider excluded the change
04

Contracts with AI vendors

Art. 13, 25(4), 26 AI Act · Art. 28 GDPR

Deployers of high-risk AI depend on the provider’s information: the instructions for use under Art. 13, details on performance limits, input data and human oversight, access to logs and a reporting path for risks and serious incidents (Art. 26(5), (6)). Without that information the deployer duties can hardly be evidenced, and the DPIA remains incomplete (Art. 26(9)).

Where the vendor is itself the provider of a high-risk system and sources components, models or services from third parties, Art. 25(4) requires a written agreement on information, capabilities, technical access and assistance – with an exception for freely licensed open-source components that are not general-purpose AI models. Alongside sits the DPA under Art. 28 GDPR with the familiar points, plus the AI-specific question of whether the vendor uses inputs for its own purposes such as model training.

  • Instructions for use, performance limits, oversight guidance
  • Log access, reporting paths, notice of changes
  • Exclude or regulate use of inputs for training
05

Evidence for authorities, customers and auditors

Art. 26(12), Art. 99(5) AI Act · Art. 5(2) GDPR

Deployers cooperate with the competent authorities in any action relating to the high-risk system (Art. 26(12)). Supplying incorrect, incomplete or misleading information to authorities risks a fine of up to €7.5 million or 1% of worldwide annual turnover (Art. 99(5)), for SMEs the lower amount. The quality of documentation is therefore a compliance topic in itself.

Robust evidence answers three questions: what was decided, on what basis, and who changed it when? Recommendation: keep classifications, approvals and duty catalogues in a system with a tamper-evident change log, not in emails and file shares. Towards customers: describe the implementation status rather than give a blanket assurance of “compliance” – that also protects against warranty and liability questions arising from your own promise.

  • Reasons, basis and change history for every decision
  • No blanket compliance promises in customer contracts
  • Answers to authorities from a single source
06

Employment law and co-determination

§ 80(3), § 87(1) no. 6, § 90(1) no. 3 BetrVG · Art. 26(7) AI Act

In Germany, an employer planning to use AI in work processes and workflows must inform the works council in good time (§ 90(1) no. 3 BetrVG). For systems suitable for monitoring employees’ behaviour or performance, a co-determination right can apply (§ 87(1) no. 6 BetrVG). Where the works council has to assess the introduction or use of AI, calling in an expert is deemed necessary to that extent (§ 80(3) BetrVG).

For high-risk AI at work the AI Act adds its own duty to inform workers’ representatives and affected workers before putting the system into service (Art. 26(7)). Both regimes run in parallel. Recommendation: a framework agreement on AI with a simplified procedure for new tools.

  • Inform early under § 90 BetrVG
  • Check co-determination where monitoring is possible
  • Observe Art. 26(7) AI Act as well

RACI excerpt

Where legal is accountable – and where it advises

Excerpt from an AI governance RACI matrix with the tasks the legal team takes part in. The highlighted column shows the role of legal and general counsel. Each task has exactly one A. The assignment is a suggestion.

Scroll the table sideways →

Where legal is accountable – and where it advises
TaskManagementAI governance leadBusiness unit / system ownerIT lead / CTOData protection (DPO)Legal
Determine the role per AI systemArt. 3, 25 AI ActIRCC–A
Classify risk and give reasonsArt. 5, 6, 50 AI ActIA/RCCCC
Assess the Art. 6(3) exceptionArt. 6(3), (4) AI ActIRCC–A
Review contracts with AI vendorsArt. 13, 25(4), 26 AI Act · Art. 28 GDPRICRCCA
Approve customer clauses and questionnairesContract practiceICR–CA
Check the FRIA dutyArt. 27 AI ActIRC–CA
Involve the works council§ 87, § 90 BetrVG · Art. 26(7) AI ActARC–IC
Answer an authority requestArt. 26(12), Art. 99(5) AI ActARCCCR
Approval decision (go/no-go)Internal processARCCCC
  • RResponsible – does the work
  • AAccountable – owns and decides
  • CConsulted – asked beforehand
  • IInformed – kept informed
  • Here the AI governance lead owns risk classification; legal is consulted and takes borderline cases such as the Art. 6(3) exception itself. In smaller organisations the two can coincide.
  • Where there is no in-house legal team, external counsel takes the C and A roles for borderline cases; routine cases stay internal.
  • The full matrix with twelve tasks including the works council is in the AI governance framework.

Interactive

The legal team’s top tasks for the next 90 days

Twelve tasks in three phases. Tick off what is done – progress shows at the top. Expand any task for details.

0 of 12 tasks done

0%

Your progress is stored in this browser only, never transmitted.

Days 1–30

0/4

Map

  • Determine the role per AI system in the inventory
  • Clarify borderline prohibited practices
  • Flag potential Annex III cases
  • Set a template for classification reasoning

Days 31–60

0/4

Contracts

  • Review existing contracts with AI vendors
  • Draft a clause module for AI procurement
  • Request vendors’ Art. 6(3) assessments
  • Customer side: align standard answer and clauses

Days 61–90

0/4

Secure

  • Check the FRIA duty under Art. 27
  • Prepare an AI framework agreement with the works council
  • Define the process for authority requests and explanation rights
  • Set a review date for the final guidelines

Order and time frame are a recommendation, not a legal requirement. Organisations with their own AI products have a much larger catalogue of duties.

From legal’s point of view

Questions that reach legal

Typical questions from sales, purchasing, business units and management – with answers you can use as a starting point.

“The customer wants us to assure AI Act compliance. Do we sign?”

Be careful with blanket promises. Many duties only apply from December 2027 and guidelines are partly in draft. Better: specific, verifiable commitments – such as keeping an AI inventory, reasoned classification, information on AI use, cooperation with requests.

“We are customising the purchased tool. Does that make us the provider?”

Only under the conditions of Art. 25(1) and only for high-risk systems: own branding, substantial modification or a purpose change that makes the system high-risk. Configuration within the intended purpose is usually not enough.

Check Annex III

“The vendor says its recruiting tool isn’t high-risk.”

Ask to see the documented assessment under Art. 6(4) and check its plausibility: which condition of para. 3 is supposed to be met? Is there profiling? Does the intended purpose match your actual use?

“Do we need to do anything for our chatbot?”

Since 2 Aug 2026 providers must design systems so that people know they are interacting with AI unless it is obvious (Art. 50(1)). As a deployer you check that the notice actually appears and observe the duties in para. 4 for AI-generated content.

Labelling duty

“What do we tell the supervisory authority?”

Only what is documented – completely and correctly. Incorrect, incomplete or misleading information is subject to fines under Art. 99(5). A central inventory with a change log is the best source.

“Do we need a fundamental rights impact assessment?”

Only if you are among the deployers in Art. 27: public bodies, private providers of public services or deployers under Annex III 5(b)/(c). For many private organisations the answer is no – but the result should be recorded with reasons.

Deadlines

The dates legal should keep in view

These dates come from the AI Act and the Digital Omnibus. Next to each is what it means for legal work.

  1. 1 Aug 2024

    AI Act enters into force

    Definitions and structure frame every classification.

  2. 2 Feb 2025

    Prohibited practices (Art. 5) and AI literacy (Art. 4)

    Assess borderline prohibitions; Art. 4 after the Omnibus as a duty to take promotion measures.

  3. 2 Aug 2026

    General application, transparency duties under Art. 50

    Review notices, labelling, exceptions.

  4. 2 Dec 2026

    New prohibitions, marking for legacy systems

    Two new prohibitions; Art. 50(2) for systems placed on the market before 2 Aug 2026.

  5. 2 Dec 2027

    High-risk duties for Annex III systems

    Art. 6, 25, 26, 27, 86 become practically relevant; contracts must fit by then.

  6. 2 Aug 2028

    High-risk duties for Annex I systems

    AI in regulated products; interplay with product safety law.

Status after the Digital Omnibus, Regulation (EU) 2026/1744, in force since 27 July 2026. All AI Act deadlines →

With SimpleAct

Classification, reasons and change history in one place

SimpleAct runs the classification through a guided questionnaire, derives the applicable duties and logs every change. The view on the right shows the risk classification with sample data from the demo tenant.

  • Risk classification

    all plans

    Guided questionnaire on high-risk triggers, transparency duties and context of use; result with Annex III category and requirements per article.

  • Legal Logic

    Professional

    Questionnaire on role, risk category and use case; the result lists per article whether it applies or not – with reasons. Review cycle and review owner with reminders.

  • Audit trail

    all plans

    Every entry with person and role, source (UI, API, system), old and new value. Built as a hash chain; integrity can be checked per record or for the whole organisation.

  • Reports and documents

    all plans

    Reports as PDF or DOCX with versioning. SimpleAct’s DPA, SLA and security whitepaper are ready for your vendor review.

Plan assignment per the pricing page. Legal Logic results are a structured decision aid and do not replace a legal review of the individual case.

app.simpleact.de

Risk Classification

AI-based Application Screening

A
Select system
Answer questions
Result

Section A: High-risk triggers

18/21 answered
2

Does the system make or influence decisions about natural persons?

YesNo
3

Does the outcome have legal or similarly significant effect?

YesNo
4

Are special categories of personal data processed?

YesNo

Risk level

High

Annex III · Employment & workforce management

Requirements

  • Risk management systemArt. 9 EU AI ActFulfilled
  • Data governanceArt. 10 EU AI ActPartial
  • Human oversightArt. 14 EU AI ActMissing

FAQ

Frequently asked questions from legal teams

Mainly map and reason: determine the organisation’s role for each AI system, review the classification including exceptions, settle duties and information in contracts with vendors and customers, and make sure decisions are traceably documented. Operational implementation sits with business units, IT and governance.
A provider develops an AI system or has it developed and places it on the market or puts it into service under its own name or trademark (Art. 3(3) AI Act). A deployer uses an AI system under its own authority, except in a personal non-professional activity (Art. 3(4)). The role is determined per system.
Under Art. 25(1) AI Act, when it puts its own name or trademark on a high-risk system already on the market, substantially modifies it so that it remains high-risk, or changes the intended purpose of an AI system so that it becomes high-risk. The initial provider must then cooperate under para. 2 unless it expressly excluded such a change.
When an Annex III system poses no significant risk to health, safety or fundamental rights and performs a narrow procedural task, improves a completed human activity, detects decision-making patterns without replacing the human assessment, or performs a preparatory task. It never applies where natural persons are profiled. The provider documents its assessment (para. 4); registration remains mandatory after the Digital Omnibus.
From a deployer’s perspective in particular: provision of the instructions for use and information under Art. 13 AI Act, details on performance limits and human oversight, access to logs, reporting paths for risks and incidents, notice of material changes and a rule on use of inputs for training. Plus the DPA under Art. 28 GDPR where personal data is processed.
Blanket assurances are risky because many duties only apply from 2 December 2027 and guidelines are partly still in draft. Specific, verifiable commitments are better: for example keeping an AI inventory, reasoned classifications, information about AI use and cooperation with requests.
Fines under Art. 99 are imposed on the operator that breaches a duty – usually the organisation in its role as provider or deployer. Civil liability and any personal responsibility of members of management depend on national law and the individual case.
Supplying incorrect, incomplete or misleading information to authorities risks up to €7.5 million or 1% of worldwide annual turnover under Art. 99(5) AI Act, whichever is higher; for SMEs the lower amount applies. For prohibited practices the ceiling is €35 million or 7%, for other duties such as Art. 26 and Art. 50 €15 million or 3%.
No. The Commission published a draft on 19 May 2026, the consultation ran until 23 July 2026, and the final version is expected by the end of 2026. Classifications should therefore be documented with assumptions and a review date.
The data processing agreement under Art. 28 GDPR with the list of sub-processors (Hetzner, Cloudflare, Stripe, Brevo, Anthropic (AI features only after opt-in; support chat see privacy policy)), the SLA and the security whitepaper with TOMs. Application and database stay in Germany at Hetzner; payments are processed via Stripe. The ISO 27001 certificate covers the data centre.

More questions? We're happy to help. Send email · Try 7 days (credit card required)

Sources and status

Last reviewed on · SimpleAct editorial team

This page is for orientation and is not legal advice. The text published in the Official Journal of the EU is authoritative. RACI assignment, checklist and time frames are recommendations, not legal requirements.

Changes to this page

  • 6 October 2026 – Rebuilt as a guide: roles under Art. 3 and Art. 25, classification and the Art. 6(3) exception, contract points, evidence, employment law, RACI excerpt, interactive 90-day checklist, ten FAQs. Blanket liability statement removed.
  • 18 May 2026 – First version as a role page for legal teams.

Every classification reasoned, every change traceable

SimpleAct connects classification, Legal Logic, approvals and the audit trail – so your legal team works from reliable data.