Zum Hauptinhalt springen
AI Compliance Starter Kit · Register

AI Risk Register

Updated on 1 min Reviewed by: Cenedril Editorial
ISO/IEC 42001 6.1.2EU AI Act Art. 9

A system for pre-screening job applications disadvantages older applicants, a chatbot states a wrong notice period, a forecasting model loses accuracy after a year. The AI risk register records such hazards per system, rates them by impact and likelihood and shows what remains after treatment.

What does the template contain?

The CSV file has 22 columns and eight example risks for the systems in the AI register, including a risk of the quality management system itself (category “Quality management system”). The scale runs from 1 to 5, the risk score is the product of impact and likelihood; levels: Low up to 5, Medium from 6, High from 12, Critical from 20.

Template: AI Risk Register

Register Template

AI Risk Register

IDHazardCategorySystem groupAI systems concernedImpact (1-5)Likelihood (1-5)Risk scoreRisk levelSeverity for affected persons (1-5)Vulnerable groups concernedMisuse coveredHarm categoriesRisk ownerResidual impactResidual likelihoodResidual risk scoreResidual risk levelAcceptedAccepted byAccepted onNote
KR-001Bias and discriminationAI systemHR processesPre-screening of job applications4312High4People with disabilities; older peopleProposal used as the sole basis for rejectionDiscrimination; loss of an employment opportunityHead of HR326MediumYesHead of HR2026-09-02Monthly analysis of proposals by group
KR-002Excessive automation (automation bias)AI systemHR processesPre-screening of job applications4312High4–Adopting the ranking without own reviewLoss of an employment opportunityHead of HR326MediumYesHead of HR2026-09-02–
KR-003Hallucination and incorrect outputsAI systemGenerative AICustomer service chatbot; writing assistant (generative AI)3412High2–Unreviewed publication of AI textsIncorrect information; financial lossHead of Customer Service236MediumYesHead of Customer Service2026-09-15Handover to a person for contract and payment questions
KR-004Attacks on confidential dataAI systemGenerative AIWriting assistant (generative AI)428Medium3–Entering confidential data into tools not approved for itDisclosure of personal dataHead of IT313LowYesHead of IT2026-08-05–
KR-005Immature technology / model driftAI systemInsurance processesRisk assessment for life and health insurance applications4312High4Older people–Unjustified rejection; financial consequencesHead of Data Science326MediumYesHead of Underwriting2026-07-01Quarterly performance check against the acceptance thresholds
KR-006Poor data qualityAI systemInsurance processesRisk assessment for life and health insurance applications428Medium4People with disabilities–Discrimination; financial consequencesHead of Data Science313LowYesHead of Underwriting2026-07-01–
KR-007Adversarial examples (targeted evasion)AI systemInsurance processesRisk assessment for life and health insurance applications326Medium2–Deliberately embellished information in the applicationFinancial loss for the organisationHead of IT Security313LowNo––Confirmation by the risk owner is pending
KR-008Failure of the external software holding the quality management system documentationQuality management systemManagement systemAll provider systems326Medium1––Loss of evidenceAI Officer212LowYesAI Officer2026-07-01Weekly backup of the documentation and a documented fallback

Sources

Frequently asked questions

Do I need a separate AI risk register next to the ISMS risk register?

The methodology can be the same, the subjects differ: the AI risk register rates hazards from AI systems for people and the organisation, including the severity for affected persons. Many organisations keep both registers with the same scale and the same risk levels.

What does the column "Severity for affected persons" mean?

It rates the consequence for the affected persons separately from the harm to the organisation, for example discrimination or the loss of an employment opportunity. Art. 9 requires a view on health, safety and fundamental rights.

How do the risk register and the impact assessment fit together?

Every foreseeable misuse from the impact assessment is assigned to a risk scenario and rated in the register. This way, the consequences for affected persons and the risks to the organisation run through the same procedure.

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.