Learning path · Level 8 · Tutorial 4

Navigating the EU AI Act's Risk-Based Rules

Use a simplified scenario to identify possible EU AI Act requirements and provider-versus-deployer responsibilities. Record what must be confirmed through current official sources and qualified advice rather than making a definitive compliance claim.

  • Level: Intermediate
  • Duration: 20 min
  • 8 steps
Start tutorial Back to the path
Navigating the EU AI Act's Risk-Based Rules

What you will go through

  1. Start with the Use and Its EU Connection Identify the facts needed to begin an EU AI Act applicability check without making a compliance claim. 2 min
  2. Read the Regulatory Map 2 min
  3. Classify the Use, Not the Product Label 3 min
  4. Identify Providers, Deployers, and Role Changes Assign provisional provider and deployer roles for HireAssist, then identify when a changed intended purpose requires reassessment. 2 min
  5. Match Possible Duties to the Responsible Role Allocate illustrative high-risk and transparency duties to the responsible role while keeping the result provisional and evidence-based. 3 min
  6. Separate the Model Layer from the System Layer 2 min
  7. Verify Sources and Application Dates Trace possible EU AI Act duties to authoritative sources, distinguish original application dates from current legal status, and record what remains unverified. 3 min
  8. Create a Dated Verification Note Produce a dated, evidence-aware verification note for Cedar Services’ proposed Applicant ranking extension without making a compliance claim. 3 min

What you will learn

  • Explain the difference between prohibited practices, high-risk systems, specified transparency duties, and lower-risk uses.
  • Distinguish provider and deployer roles in a scenario and identify changes that could require reassessing those roles.
  • Identify possible obligations while separating general-purpose model responsibilities from responsibilities for a deployed system.
  • Create a dated verification note recording the relevant official sources, application dates, unresolved questions, and reasons to seek qualified advice.

Before you start

  • Identifying AI Tasks and the Products Around Models
  • Assessing Bias with Contrasting Cases
  • Creating an AI Privacy and Security Checklist
  • Checking AI Content Rights and Authenticity

Step 1 of 8

Start with the Use and Its EU Connection

Identify the facts needed to begin an EU AI Act applicability check without making a compliance claim.

Begin with the use

Start with a fact pattern

Begin an EU AI Act check with a fact pattern, not a product label or a claim that a system is compliant.

Capture four things:

  1. Intended purpose — what the AI system is meant to do.
  2. Actual use — what people have configured and use it to do in practice.
  3. Affected people and decision influence — who receives, relies on, or may be affected by its outputs, and whether outputs influence decisions.
  4. EU connection — relevant placement on the market, use, or—in specified circumstances—use of outputs in the EU.

These facts support educational triage: identifying possible requirements and questions to escalate. They do not establish a legal compliance conclusion.

Map the facts before judging them

A diagram maps Harbor AI, Cedar Services in Ireland, prospective applicants, and the HireAssist system used for FAQ answers.

The diagram is a fact map, not a legal classification. It separates organizations, the AI system, affected people, and the use to investigate.

Separate known facts from missing facts

Example

Known scenario facts versus evidence to seek

The running scenario is deliberately narrow:

Cedar Services is an employer established in Ireland. It uses the AI system HireAssist for FAQ chat: answering questions about published vacancies and the application process. The stated baseline does not evaluate applicants or rank them.

Harbor AI develops and markets HireAssist under its own name. Those facts matter, but an organization’s address and a label such as “chatbot” do not settle whether or how the EU AI Act applies.

Compare the stated purpose with evidence of the live configuration and real workflow. A system can be described one way yet used more broadly.

A first fact inventory

Use this short inventory to keep established facts separate from facts that still need confirmation.

Record uncertainty rather than filling gaps with assumptions.
AreaEstablished in the scenarioStill to confirm
Intended purposeFAQ chat about vacancies and the application processThe documented intended purpose and any stated limits
Actual useNo applicant evaluation or ranking is describedWhether the live system and staff workflow remain FAQ-only
Affected peoplePeople seeking information about vacanciesWhich user groups access it and whether others are indirectly affected
Decision influenceFAQ answers are described, not hiring evaluationWhether outputs influence applicant screening or hiring decisions
EU connectionCedar Services is established in IrelandRelevant market placement, use locations, and any output use in the EU

Keep scope questions open

Check the EU connection, then the scope

Article 2 of Regulation (EU) 2024/1689 is a starting point for scope, together with relevant definitions and exclusions. A useful first question is whether there is a relevant EU connection—for example, an AI system placed on the market, put into service, or used in the EU.

Location alone is not decisive. The scope can also matter when an organization is outside the EU but outputs produced by an AI system are used in the EU, in the circumstances specified by the legal text.

For Cedar Services, its establishment in Ireland is an important fact to record. It is not, by itself, a complete applicability analysis. Confirm the system’s route to market, deployment context, actual use, and any relevant exclusions from current official sources.

Tip

Do not turn triage into a compliance claim

Keep this exercise educational and evidence-aware. The EU AI Act operates alongside privacy, copyright, employment, and other sector-specific requirements. Existing safeguards remain relevant; they do not substitute for checking the applicable legal text and obtaining qualified advice when the facts or consequences warrant it.

Practice: record the starting facts

Create a first triage note

Using the FAQ chat scenario, write four brief bullets:

  • its intended purpose;
  • the people affected;
  • the EU connection you can identify;
  • one fact that still needs confirmation before an applicability check can go further.

Write at least 80 characters (0/80).

Step 2 of 8

Read the Regulatory Map

See the map before classifying

Start with the treatment map

A map of treatments—not a score for “smartness”

The EU AI Act’s risk-based approach asks what an AI system is used for, its likely impacts, and whether defined legal conditions are met. It does not sort systems simply by model capability, output quality, or an AI product label.

For example, HireAssist’s FAQ chat may be useful or inaccurate, but neither quality alone determines its Possible treatment. The use, context, and legal conditions must be examined.

Use this map for educational triage, not as a compliance conclusion.

How the layers relate

Diagram showing Cedar Services as an Organization, HireAssist as a System, and BaseText as a Model, with four possible regulatory treatments assessed around the system.

A treatment follows the use and applicable legal conditions; it is not a rating of the model’s intelligence.

Read each label precisely

Four meanings to keep separate

The four labels below are an orientation tool. They should not be treated as four sealed boxes or as a final legal result.

Possible treatment labels for an initial EU AI Act triage
Treatment labelCore consequenceWhat it does not mean
Prohibited practicesThe practice cannot proceed when the legal prohibition applies.Disclosure or human review makes it acceptable.
High-risk systemsThe system is subject to specified requirements.The system is automatically prohibited.
Specified transparency dutiesDefined interactions or synthetic-content situations can trigger information or disclosure duties.Every AI-assisted output needs a label.
Lower-risk usesAn explanatory shorthand for uses outside those treatments in the simplified map.No EU AI Act duties or other laws can apply.

Overlap is possible

A system can have more than one relevant treatment. In particular, a high-risk AI system can also have a specified transparency duty.

Rules concerning a general-purpose AI model are another layer. A model such as BaseText may support systems with very different uses. Recognizing that separate layer does not decide the treatment of HireAssist.

Tip

Keep the claim narrow

Do not treat lower-risk as risk-free. The label is useful shorthand, but privacy, copyright, employment, consumer-protection, and other applicable rules may still matter.

Practice: test an overlap claim

Check the overlap rule

True or false: If HireAssist is subject to a specified transparency duty, it cannot also be a high-risk AI system.

Write at least 4 characters (0/4).

Practice: complete the map

Apply the regulatory map

Match each treatment label to the consequence that best describes it.

Match treatment to consequence

Match each treatment to its regulatory consequence.

Tap an item and then its matching pair.

Step 3 of 8

Classify the Use, Not the Product Label

Begin with the system’s function

Classify the use, not the label

A product label such as “chatbot” or “employment AI” is not a legal classification. Start with the AI system’s intended purpose, what it actually does, who is affected, and how its output affects a decision.

FAQ chat is HireAssist answering questions about published vacancies and the application process. It does not evaluate applicants.

The proposed Applicant ranking extension evaluates applications and ranks candidates for hiring decisions. That changed function—not the system’s name—is the important starting fact.

Use Possible treatment for a provisional result. Definitions, exceptions, evidence, and current official sources still need checking.

Different functions, different starting facts

Comparison of HireAssist providing vacancy information and HireAssist ranking applications for a hiring decision.

The decisive difference is whether the AI system evaluates people for a hiring decision.

Warning

Keep the conclusion provisional

This is educational triage, not a compliance determination. The EU AI Act must be checked alongside other applicable requirements and current official sources.

Follow the Article 6 routes

Locate the possible high-risk route

Article 6 provides two main high-risk routes.

Article 6(1) concerns an AI system that is a product, or a safety component of a product, covered by specified Union harmonisation law in Annex I—and that must undergo a third-party conformity assessment under that law.

Article 6(2) concerns AI systems with an intended purpose listed in Annex III. For Applicant ranking, the relevant starting point is Annex III’s employment provisions, because applications are evaluated and candidates are ranked for hiring.

This is a starting analysis, not a final classification.

Two Article 6 routes and a conditional check

A flow diagram showing the Article 6(1) product or safety-component route and the Article 6(2) Annex III route, with Article 6(3) shown as a conditional, justified check rather than an automatic exemption.

Article 6(3) is a limited conditional assessment within the Annex III route; it is not an automatic exemption.

Treat Article 6(3) as a justified exception

Article 6(3) contains limited conditions under which an Annex III use may not be high-risk because it does not pose a significant risk to health, safety, or fundamental rights. Those conditions require justification; they are not created by simply adding a human reviewer.

An Annex III use involving profiling of natural persons remains high-risk. The scenario does not provide enough facts to rely on an Article 6(3) exception. Confirm the intended purpose, decision influence, system features, and current legal text before reaching a conclusion.

Article 6 triage map

Use the route and conditions together; neither route is determined by a broad sector label alone.

Possible starting routes for the scenario; current provisions and facts require verification.
ProvisionStarting questionScenario signalWhat still needs checking
Article 6(1)Is it a specified regulated product or safety component requiring third-party assessment?No such product facts are given.Product status and applicable Annex I law.
Article 6(2) and Annex IIIIs the intended purpose a listed use?Ranking applications for hiring is the relevant employment starting point.Exact function and decision influence.
Article 6(3)Could a limited Annex III exception be justified?Human review alone is insufficient.Risk, profiling, justification, and legal conditions.

Recognize a prohibition boundary

Check the Article 5 boundary

A separate boundary check applies when an AI system infers emotions in the workplace from biometric data for productivity monitoring. Do not generalize from the word “emotion” alone.

Article 5 includes a prohibition relevant to emotion inference in workplaces, while recognizing an exception where the system is intended for medical or safety reasons. For a productivity-monitoring proposal, record a Possible treatment of Prohibited practices and check the Article 5 scope and whether any exception genuinely applies.

A beneficial claim, disclosure, or human review does not itself answer that Article 5 question.

A purpose-sensitive prohibition boundary

Comparison of workplace emotion inference for productivity monitoring and a possible medical or safety purpose, showing that the Article 5 scope and exception need checking.

The workplace purpose is decisive. Do not assume that every emotion-related system has the same treatment.

Practice: identify the boundary

Choose the appropriate boundary check

Cedar Services proposes a tool that infers workers’ emotions from biometric data for productivity monitoring. What is the sound provisional next step?

Final practice: distinguish the three uses

Apply the classification triage

Write a short triage note covering all three uses:

  1. FAQ chat
  2. Applicant ranking
  3. Workplace emotion inference for productivity monitoring

For each, state a Possible treatment, name the decisive fact, and identify any exception or unresolved point that must be checked. Keep the result provisional rather than claiming compliance.

Write at least 300 characters (0/300).

Step 4 of 8

Identify Providers, Deployers, and Role Changes

Assign provisional provider and deployer roles for HireAssist, then identify when a changed intended purpose requires reassessment.

Start with actions, not labels

Two system-level roles

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—subject to the legal definition.

A deployer uses an AI system under its authority. Personal, non-professional use is excluded from that definition.

These are role labels, not rankings. Organization size, subscription price, or who pays the invoice do not decide the role. Start with what each organization does with the AI system.

Follow the relationship to the system

Diagram showing Organization Harbor AI connected to System HireAssist as its developer and marketer, and Organization Cedar Services connected to HireAssist as its user.

Map the actions around the AI system before assigning provisional roles.

Assign provisional baseline roles

Map the baseline scenario

In the baseline use, HireAssist provides FAQ chat: it answers questions about published vacancies and the application process. It does not evaluate applicants or rank them.

This is a provisional, system-level map for educational triage. Confirm the actual arrangements, documentation, and applicable definitions before relying on it.

Possible baseline role map

The same organization can hold more than one role in a different arrangement.

Roles follow the organization’s relationship to the AI system, not its size or payment arrangement.
OrganizationRelationship to HireAssistPossible role
Harbor AIDevelops and markets HireAssist under its own nameprovider
Cedar ServicesUses HireAssist for recruitment FAQ chatdeployer

Reassess after a material change

Recognize a reassessment trigger

Roles must be reassessed when the relevant facts change. One organization can hold multiple roles, and an AI system developed for internal use can still raise provider responsibilities.

For high-risk AI systems, Article 25 identifies circumstances that can shift provider responsibilities. They include specified rebranding, substantial modification, and changing a previously non-high-risk system’s intended purpose so that it becomes high-risk.

Ordinary configuration does not automatically turn a deployer into a provider. But a material change of purpose cannot be dismissed because the original supplier label remains unchanged.

A changed purpose requires a checkpoint

Flow diagram from HireAssist FAQ chat to proposed applicant ranking, ending at a conditional Article 25 reassessment checkpoint for possible provider responsibilities.

The proposed use change calls for reassessment; it does not automatically transfer the provider role.

Example

Apply the change test

Cedar Services proposes Applicant ranking: HireAssist would evaluate applications and rank candidates for hiring decisions. This is outside Harbor AI’s stated FAQ-only intended purpose.

The earlier treatment check makes this consequential employment use a reason for fresh assessment. Record the intended-purpose documentation, who made the change, what was modified, and whether the Article 25 conditions apply. Do not make a definitive role or compliance claim from the product name alone.

Apply the baseline map and change test

Check the role change

Which statement best records the baseline roles and the appropriate response to Cedar Services’ proposed Applicant ranking extension?

Step 5 of 8

Match Possible Duties to the Responsible Role

Allocate illustrative high-risk and transparency duties to the responsible role while keeping the result provisional and evidence-based.

Map possible high-risk responsibilities

Possible high-risk duties follow the role

For the proposed Applicant ranking use, first confirm whether HireAssist is a high-risk AI system. If it is, duties are allocated by role—not by who paid for the tool.

Possible treatment: Harbor AI is the system provider because it develops and markets HireAssist under its own name. Cedar Services is the deployer when it uses HireAssist for hiring. This is a provisional scenario mapping, not a compliance determination.

A high-risk provider may need a risk-management process, relevant data governance, technical documentation, instructions for deployers, conformity assessment, and post-market monitoring. Conformity assessment is a defined compliance process; it does not automatically mean every case needs an external certification service.

A high-risk deployer may need to follow instructions, assign competent human oversight, monitor operation, retain relevant logs under its control, and act when it identifies risks or serious incidents.

A diagram separates Harbor AI's provider-side responsibilities from Cedar Services' deployer-side responsibilities around the HireAssist system.

The same AI system can create different possible duties for its provider and deployer.

Role-to-duty evidence map

Use this as an evidence-request map, not as a complete checklist or proof that Regulation (EU) 2024/1689 applies to the scenario.

Illustrative duties if Applicant ranking is confirmed as high-risk.
RolePossible dutiesEvidence to request
Provider: Harbor AIRisk management; data governance; documentation; instructions; conformity assessment; post-market monitoringIntended-purpose documents, instructions, technical documentation, assessment records
Deployer: Cedar ServicesInstructed use; competent oversight; monitoring; relevant controlled logs; response to risks or serious incidentsOversight assignment, operating procedure, monitoring records, log-access process

Add transparency and AI literacy

Transparency and literacy are separate duty layers

High-risk duties are not the only possible layer. Article 50 includes specified transparency duties that depend on the system's function and context.

For an AI system intended to interact directly with people, the provider has an information duty so people are informed that they are interacting with AI, subject to conditions including whether that interaction is obvious from the circumstances and context. A notice is therefore not a universal substitute for every other duty.

For specified synthetic outputs, provider-side machine-readable marking and deployer-side disclosure of deepfakes are different possible responsibilities. Their conditions, exceptions, and the facts of a particular output must be checked in the current legal text.

AI literacy is another cross-cutting duty: in-scope providers and deployers should take measures for a sufficient level of AI literacy among people dealing with AI on their behalf. It is not limited to high-risk systems.

Specified transparency duties: a role split

Separate the output-related roles before deciding what evidence to request.

Illustrative Article 50 distinctions; confirm current conditions and exceptions.
SituationPossible responsible rolePossible action to examine
Direct AI interaction with a personProviderInformation that the person is interacting with AI, where Article 50 conditions apply
Specified synthetic outputProviderMachine-readable marking of output as artificially generated or manipulated
Deepfake use or disseminationDeployerDisclosure that the content has been artificially generated or manipulated
Staff use of an AI systemProvider and deployerAppropriate AI literacy measures for relevant people

Warning

Keep the conclusion provisional

Do not treat a chatbot notice, a human reviewer, or a short role-to-duty table as a complete compliance result. Possible duties can overlap, and applicability depends on the use, legal conditions, current text, and evidence.

Check the role allocation

Practice: assign the possible duty

For the proposed Applicant ranking use, match each possible duty to the most relevant role.

Tap an item and then its matching pair.

Apply the role-to-duty map

Apply the duty map

Write a short provisional duty note for the proposed Applicant ranking extension.

  1. Allocate these possible duties to provider, deployer, or both: system-level technical documentation; use according to instructions with competent human oversight and relevant controlled logs; AI literacy measures.
  2. Explain why a chatbot notice or a human reviewer alone would not satisfy all potentially relevant requirements.

Use conditional language such as “possible” and “if confirmed high-risk.”

Write at least 220 characters (0/220).

Step 6 of 8

Separate the Model Layer from the System Layer

Trace the layers before assigning duties

One model, different downstream uses

A general-purpose AI model can be incorporated into many downstream products. After introducing that term, it is often shortened to general-purpose AI (GPAI).

Keep three layers separate in this scenario:

  • Model: Meridian Models provides BaseText.
  • System: Harbor AI develops and markets HireAssist under its own name.
  • Organization and use: Cedar Services uses HireAssist in Ireland.

The same BaseText model may support HireAssist for FAQ chat and, if the extension proceeds, Applicant ranking. Those uses can have different Possible treatments. A model’s capabilities alone do not settle the treatment of a particular AI system or its use.

Model-to-system-to-use map

A layered diagram maps Meridian Models and BaseText to Harbor AI and HireAssist, then to Cedar Services using FAQ chat or proposed Applicant ranking.

The Model, the AI system, and the Organization’s use are connected—but they are not the same regulatory layer.

Map possible duties to the right layer

Do not transfer duties across layers

Possible duties should be mapped to the actor and layer they concern. This is an educational triage map, not a conclusion that any actor has met every requirement.

Possible responsibilities by layer

Illustrative responsibilities to investigate for the scenario

Confirm the current legal text, applicable conditions, and evidence with official sources and qualified advice.
LayerActor and objectPossible focus
ModelMeridian Models / BaseTextTechnical documentation; information for downstream system providers; copyright-compliance policy; sufficiently detailed public summary of training content.
SystemHarbor AI / HireAssistSystem-provider responsibilities relevant to its intended purpose and Possible treatment, including any consequences of Applicant ranking.
UseCedar Services / Applicant rankingDeployer responsibilities connected to use under its authority, including duties previously mapped for a high-risk use if that treatment applies.

Tip

A separate model-level classification

A GPAI model with systemic risk can face additional model-level obligations. That is not the same classification as a high-risk AI system. Do not infer either conclusion from the other, and do not assume that Meridian Models’ possible model-level duties replace Harbor AI’s or Cedar Services’ system-level duties.

Separate two different classifications

Check the distinction

True or false: A GPAI model with systemic risk and a high-risk AI system are the same regulatory classification.

Apply the layered map

Map the actors and their possible duty layers

Match each organization to its place in the model-to-system-to-use chain and the duty layer to assess for the proposed Applicant ranking extension.

Tap an item and then its matching pair.

Step 7 of 8

Verify Sources and Application Dates

Trace possible EU AI Act duties to authoritative sources, distinguish original application dates from current legal status, and record what remains unverified.

Choose and describe authoritative sources

Start with the source type

Begin with authority, then check currency. Use EUR-Lex to read Regulation (EU) 2024/1689, identify the relevant Article or Annex, and check the version and any applicable amendments. That legal text is the starting point for a possible duty.

European Commission and AI Office materials can provide official implementation guidance. They can help explain a process, but guidance does not replace a binding legal provision. A legislative proposal describes a possible future change; it is not an enacted change.

For every source, record its title, specific provision or topic, version, and the date you accessed it.

A source hierarchy for verification

A diagram showing binding legal text, official guidance, and a legislative proposal flowing into a dated verification record.

Identify what a source is before relying on it. The enacted legal text is the source of a binding provision.

Tip

Record access separately from legal effect

Use Checked on for the date you accessed a source. It records access; it does not prove the source will remain current. A later amendment, consolidated version, or transitional rule may change what needs assessment.

Read the schedule, then verify the date

Application dates need provision-level checking

Do not turn the schedule into one generic deadline. Entry into force and application are different questions: entry into force concerns when the Regulation became legally effective, while application asks when a particular provision and its obligations apply.

The table and timeline below show the Original enacted schedule. Use them only as a historical starting point. Then confirm the current provision, amendments, and transitional rules that apply to the facts.

A date is a starting point, not the answer

A timeline of four Original enacted schedule milestones leading to a separate checkpoint for current provisions, amendments, transitions, and lifecycle dates.

Use the historical schedule to frame a question, then verify the provision-specific date against current law and the system’s lifecycle facts.

Original enacted schedule

Historical starting point only. Confirm the current legal text and relevant transitional rules before relying on a date.

Original enacted schedule for Regulation (EU) 2024/1689
AreaStarting dateWhat to verify
Prohibitions and AI literacy2025-02-02Relevant Article and scope
GPAI provisions2025-08-02Transitions and applicable model provision
Most provisions2026-08-02Specific duty and any amendment
Article 6(1) and corresponding obligations2027-08-02Product or safety-component route and transition

Connect the date question to the lifecycle

Ask a date question that fits the facts: “Which application date and transitional rule apply to this possible duty?” Record when HireAssist was placed on the market or put into service, and when a relevant change of purpose occurred.

Do not assume that FAQ chat and the proposed Applicant ranking extension share one deadline simply because they use the same AI system name.

Keep unknowns visible in the record

Example

Illustrative record fragments

Sources and status
Practice entry: EUR-Lex, Regulation (EU) 2024/1689, Article 50; version shown on the source page; Checked on: 2026-05-14; status: Checked.

Application dates and transitions
Possible duty: information for people interacting with an AI system. Confirm the current application rule for the specific provision; do not infer it from a general schedule.

Open questions
When was HireAssist put into service? When would Applicant ranking begin? Has a change in intended purpose been documented?

This illustrates an evidence record. It does not verify today’s law.

Warning

Do not promote an unread source to evidence

If a source cannot be opened or read, label it Not yet checked, not Checked. A dated illustrative extract can support practice, but it does not establish that current legal text has been verified. Use Needs clarification for a material claim or fact that remains uncertain and needs follow-up.

Practice: verify without overclaiming

Practice: verify without overclaiming

Use the dated training records below to draft five short bullets.

  • Record A: EUR-Lex display of Regulation (EU) 2024/1689, Article 113; source page accessed 2026-05-14.
  • Record B: European Commission AI Office webpage, “Implementation overview”; source page accessed 2026-05-14.
  • Record C: European Commission document titled “Proposal for a Regulation amending rules on AI”; source page accessed 2026-05-14.
  • Record D: An inaccessible saved link said to contain a later official update.
  1. Classify A, B, and C by legal weight.
  2. Give the provision-specific application-date and transition question to investigate before relying on a date for Applicant ranking.
  3. Assign the appropriate source-status label to D.
  4. An undated internal slide says, “Applicant ranking has one generic 2026-08-02 deadline.” Assign the appropriate source-status label and state why.
  5. Name one lifecycle fact that must be established before transition analysis can proceed.

Write at least 240 characters (0/240).

Step 8 of 8

Create a Dated Verification Note

Produce a dated, evidence-aware verification note for Cedar Services’ proposed Applicant ranking extension without making a compliance claim.

1. Define the changed scenario

Turn analysis into a record

A verification note is a dated triage record—not a compliance certificate. It separates three things:

  • Established facts: what the organization can evidence now.
  • Possible treatment and duties: provisional conclusions tied to provisions that still need checking.
  • Open questions: facts, legal conditions, sources, or dates that could change the conclusion.

For this scenario, keep the two uses distinct. Cedar Services continues using HireAssist for FAQ chat about published vacancies and applications. It proposes Applicant ranking, which evaluates applications and ranks candidates for hiring decisions. Harbor AI markets HireAssist under its own name with an FAQ-only stated intended purpose. Meridian Models provides the general-purpose AI model BaseText within HireAssist.

Keep the layers and uses separate

Diagram separating Cedar Services' FAQ chat and proposed applicant ranking use of HireAssist, Harbor AI's system role and documentation, and Meridian Models' BaseText model layer.

The proposed use can change the analysis even while the FAQ chat remains available. Record the system, model, organizations, evidence, and unanswered questions separately.

2. Draft without overclaiming

A note has an audit trail

Use the required headings below as a compact drafting frame. A useful note gives each possible conclusion a reason to verify it; it does not turn a short checklist into proof of compliance.

Verification-note frame

Draft each row briefly. Add source title, provision, version, and a Checked on date only after you actually access the source.

Use all headings in the final note.
HeadingWhat to record
As-of dateDate of the note, in YYYY-MM-DD format
Use and EU connectionBaseline and changed use; relevant EU facts
Possible treatmentProvisional treatment and provisions to verify
RolesSystem and model actors; possible role reassessment
Possible dutiesRole-linked duties or evidence to investigate
Sources and statusOfficial source, provision, status, and Checked on date
Application dates and transitionsOriginal enacted schedule plus current-law questions
Open questionsMissing facts and the evidence needed
Escalation and next actionProportionate action before expanded use proceeds

Example

Use evidence-aware language

Write facts, findings, and questions differently.

Established fact: Harbor AI states an FAQ-only intended purpose for HireAssist.

Possible finding: Using HireAssist for Applicant ranking may require reassessing Cedar Services’ role under Article 25 if the legal conditions are met.

Open question: Has Cedar Services changed the system’s intended purpose in a way that triggers that provision, and when would the changed system be put into service?

Use cautious words such as may, possible, subject to confirmation, and needs clarification. A statement that an official page will be consulted belongs under Not yet checked until it has actually been accessed and recorded.

3. Apply the full triage sequence

Final application: write the verification note

Create a concise verification note for this self-contained exercise scenario.

Exercise date: 2025-09-01. Cedar Services is established in Ireland. It uses Harbor AI’s HireAssist for FAQ chat about published vacancies and the application process. Harbor AI markets HireAssist under its own name and documents an FAQ-only intended purpose. Cedar proposes Applicant ranking: HireAssist would evaluate applications and rank candidates for hiring decisions, while FAQ chat remains available. HireAssist uses Meridian Models’ general-purpose AI model BaseText. No official legal or guidance source has been accessed for this exercise, and the dates of market placement, putting into service, and the proposed change are unknown.

Use every required verification-note heading. Identify the likely high-risk issue, a possible provider-role reassessment, continuing transparency considerations, and the separate GPAI model layer. Record official sources and their accurate status; identify application-date and transition questions; then state why qualified review is needed before the expanded use proceeds.

Write at least 700 characters (0/700).

4. Record, verify, and escalate

Summary

Reusable sequence

  • Start with evidenced facts about the use, affected decisions, EU connection, and intended purpose.
  • Keep FAQ chat and Applicant ranking distinct when recording Possible treatment.
  • Separate the AI system roles from the general-purpose AI model layer.
  • Mark an unaccessed source Not yet checked; record Checked on only after access.
  • Treat the Original enacted schedule as a historical starting point and verify current provisions, amendments, transitions, and relevant dates.
  • Escalate material uncertainty before the expanded use proceeds.

Step complete

Congratulations! You completed: Navigating the EU AI Act's Risk-Based Rules

You created an evidence-aware triage note: establish facts, identify possible treatments and roles, map possible duties, verify current sources and dates, then escalate uncertainty.

20 XP

Download the App now to have access to + 5000 free courses, exercises, certificates and lots of content without paying anything!

  • 100% free online courses from start to finish

    Thousands of online courses in video, ebooks and audiobooks.

  • More than 60 thousand free exercises

    To test your knowledge during online courses

  • Valid free Digital Certificate with QR Code

    Generated directly from your cell phone's photo gallery and sent to your email

Cursa app on the ebook screen, the video course screen and the course exercises screen, plus the course completion certificate