Zum Hauptinhalt springen
AI Compliance Starter Kit · Register

AI Register

Updated on 2 min Reviewed by: Cenedril Editorial
ISO/IEC 42001 A.4, A.10.2EU AI Act Art. 6, 26, 49

Anyone who cannot answer “Which AI do you use?” within a few minutes fails at the first question of every ISO/IEC 42001 audit and every request from a market surveillance authority. The AI register lists each AI system with its role, risk class, applicable obligations and the details the standard and the Regulation require per system.

What does the template contain?

The CSV file has 34 columns and five example systems of an example organisation: a customer service chatbot, a system for pre-screening job applications (high risk), a writing assistant, a document classification with an exemption under Art. 6(3) and an in-house system for risk assessment in life and health insurance (provider and deployer, high risk, fundamental rights impact assessment as a legal duty).

Template: AI Register

Register Template

AI Register

AI systemVendorRoleEffective risk classComputed risk classJustification of adjustmentClassified onResponsible personLocation of instructionsFundamental rights impact assessment requiredExemption under Art. 6(3)Condition of the exemptionJustification of the non-high-risk assessmentEU database registration (number, date)Annex III areasProhibited practices (Art. 5)Transparency obligations (Art. 50)Number of applicable obligationsApplicable obligationsObligations with implementation statusObligations approved onObligations approved byIntended purpose and area of usePerson exercising oversightParties per life cycle stageComputing and system resources per stageMetrics (value, date measured)Support channel and response timeLogging (events, access, location, retention)Information of affected persons (channel, location)Impact assessment last carried outNotification of the fundamental rights impact assessment (date, reference or exemption)Substantial modifications (date, description)Orders by authorities (authority, reference, deadline)
Customer service chatbotSaaS vendor (chatbot platform)DeployerLimited riskLimited risk–2026-09-10Head of Customer ServiceVendor's terms of use and manual, Customer Service folderNoNo––––NoneInteraction with an AI system (Art. 50(1))2Art. 4: AI literacy | Art. 50: notice of interaction with AIArt. 4: Done | Art. 50: Done (Evidence: notice text in the chat window)2026-09-15AI OfficerAnswering frequent customer questions on the website; handover to a person on requestCustomer Service team leadModel and operation: vendor; knowledge base and monitoring: Customer ServiceOperation: vendor's cloud service, EU regionShare of conversations handed over to a person 18% (2026-10-01)Ticket to the vendor, response within one working dayConversation logs at the vendor, access Customer Service, 90 days–2026-09-12–––
Pre-screening of job applicationsRecruiting software vendorDeployerHigh riskHigh risk–2026-08-20Head of HRVendor's instructions for use (version 4.2), HR folderNoNo–––Annex III point 4(a) (employment, screening of applications)None–5Art. 4: AI literacy | Art. 26(1): use in line with instructions | Art. 26(2): human oversight | Art. 26(6): logs | Art. 26(11): information of applicantsArt. 4: Done | Art. 26(1): Done | Art. 26(2): Done | Art. 26(6): Open (Responsible: IT Operations) | Art. 26(11): Done (Evidence: privacy notice in the application portal)2026-09-02AI OfficerRanking incoming applications against the job requirements; HR makes the decisionRecruiter responsible for the positionModel: vendor; input data and oversight: HR; operation: vendorOperation: vendor's cloud service, EU regionDeviation of proposals from the HR decision 9% (2026-10-01)Vendor's support, response within four hoursUse, inputs, proposals, interventions; access HR and data protection; at the vendor; 6 monthsPrivacy notice in the application portal, section on selection2026-08-25–––
Writing assistant (generative AI)AI assistant vendor (enterprise version)DeployerMinimal riskMinimal risk–2026-07-30Head of ITVendor's terms of use, intranetNoNo––––NoneAI-generated text on matters of public interest (Art. 50(4)), only when published2Art. 4: AI literacy | Art. 50(4): disclosure when publishedArt. 4: Done | Art. 50(4): Done (Evidence: approval of published content)2026-08-05AI OfficerDrafts, summaries and translations in all departments; no decisions about persons–Model and operation: vendor; use: all employeesOperation: vendor's cloud service, EU regionCompleted training 94% (2026-10-01)IT service desk, response within one working dayUsage logs at the vendor, access IT, 30 days–2026-08-01–––
Document classification for personnel filesIn-house developmentProvider and deployerMinimal riskMinimal risk–2026-09-05Head of HRInstructions for use in the development wikiNoYesThe system performs a narrow procedural task (Art. 6(3)(a))Assigns incoming documents to the sections of the personnel file; does not assess persons or influence any decision[REGISTRATION_NUMBER], 2026-09-08Annex III point 4 (employment), exemption under Art. 6(3)None–3Art. 4: AI literacy | Art. 6(4): documentation of the assessment | Art. 49(2): registrationArt. 4: Done | Art. 6(4): Done | Art. 49(2): Done2026-09-08AI OfficerFiling incoming documents in the personnel fileHR administrationDevelopment, operation and monitoring: own organisationTraining: local workstation with GPU; operation: own server in the data centreAccuracy of filing 97% (2026-09-30)Ticket to development, response within one working dayAssignments and corrections; access development; central log management; 12 months–2026-09-04–––
Risk assessment for life and health insurance applicationsIn-house developmentProvider and deployerHigh riskHigh risk–2026-06-15Head of UnderwritingInstructions for use (version 1.3), technical documentation chapter 1YesNo––[REGISTRATION_NUMBER], 2026-07-01Annex III point 5(c) (risk assessment in life and health insurance)None–6Art. 9 to 15: requirements for high-risk systems | Art. 17: quality management system | Art. 43: conformity assessment | Art. 49(1): registration | Art. 26: deployer obligations | Art. 27: fundamental rights impact assessmentArt. 9 to 15: Done | Art. 17: Done | Art. 43: Done (Evidence: EU declaration of conformity of 2026-06-28) | Art. 49(1): Done | Art. 26: Done | Art. 27: Done2026-07-02AI OfficerProposal of a risk rating for applications; Underwriting makes the decisionUnderwriter of the applicationData and model: Data Science; operation: IT Operations; oversight: UnderwritingTraining: cloud GPU, EU region; operation: Kubernetes cluster in own data centreDifference in rejection rate between age groups 1.4 percentage points (2026-09-30)Data Science on call, response within two hoursInputs, outputs, interventions, model version; access Data Science and Internal Audit; central log management; 10 yearsApplication documents, section on the use of an AI system2026-06-20Notified on 2026-07-03, reference [NOTIFICATION_REFERENCE]2026-09-10: new model after retraining, new conformity assessment–

Sources

Frequently asked questions

Which systems belong in the AI register?

Every AI system the organisation selects, uses, develops or offers, including bought-in software with embedded AI features and AI services employees use for work. Reported shadow AI is added as soon as it continues under approved conditions.

Is the AI register the same as registration in the EU database?

The AI register is the internal list of all systems. Only certain systems are entered in the EU database under Art. 49, above all providers' high-risk systems and systems with an exemption under Art. 6(3). The register records the registration number and date where there is one.

Why does the template have so many columns?

The first 22 columns carry classification and obligations. The remaining twelve are required per system by ISO/IEC 42001 and the AI Act, such as oversight, resources, logging and the information of affected persons. For a minimal-risk system many of them stay empty.

No Cookies!

This wiki collects nothing, bakes nothing and leaves nothing behind. There's nothing to consent to. Privacy doesn't get better than this.