In the AI Compliance Assistant you assess the risks of your AI systems in the chapter Risks & Impact Assessment across six consecutive pages: you create risk scenarios per AI type, rate every scenario for the organisation and for the people affected, check the calculated risk levels, set the ISO/IEC 42001 Annex A controls per system, rate the residual risk and complete the impact assessment for high-risk systems and systems in your own provider role. At the end, the Risk register (AI), the Statement of Applicability (AI), the Risk Treatment Plan and the impact assessment per system are stored as records in Cenedril.
Open the chapter
Open the Risks & Impact Assessment chapter
In the sidebar open Wizards and pick the Artificial Intelligence card. The overview AI Compliance Assistant (AIMS) shows the chapters as a list; clicking the row Risks & Impact Assessment opens the chapter entry page. It explains why the risk assessment is the core of the AI management system and previews the three steps of the chapter: Risk identification, Analysis & evaluation and Treatment & impact assessment.
Move to the first input page
Click Next at the bottom. The progress strip at the top of every page lists the steps of the chapter: AI Risks, Risk Analysis, Risk Evaluation, Risk Treatment, Treatment Effects and Impact assessment. You can reach any step directly from the strip.
Create risk scenarios
Go through the groups per AI type
The Risk identification page groups your AI systems by the AI type from the inventory, for example Generative AI (LLM, image, video, audio) or Recruitment & HR. Each group is a collapsible card listing its systems with their risk class; the header counts the scenarios. Systems of the type Other stand alone with the full catalogue. If no systems have been recorded yet, the link Go to AI inventory leads to the inventory.
Add threats as scenarios
Under Recommended threats you find the threats that fit the AI type; Other threats unfolds the rest of the catalogue. Threats that require access to training and model are recommended only in groups with a system in the provider role, but remain selectable everywhere. Every click on a chip creates a scenario for all systems of the group; the trash icon Remove scenario takes it out again. Include what a plausible story of harm can be told about for this type, and leave no group without a scenario.
Add misuse patterns and custom scenarios
The block Reasonably foreseeable misuse lists patterns of people using a system differently from how it was intended, for example Adopting outputs unchecked. Every pattern you pick becomes a scenario of its own. Via Custom scenario you record a risk outside both catalogues with a name and a Description; without a name the page cannot be completed with Save & Continue. A custom scenario is linked to no Annex A control; on the treatment effects page its list of mitigating controls stays empty.
Enrich a scenario
Unfold a scenario. Under Responsible person choose who can actually decide about the use of the system; only scenarios with this entry get the confirmation checkbox for the residual risk later. Affected AI objectives and Affected context topics are optional and draw on the AI objectives and the organisational context. If a group covers several systems, use Applies to these systems to take out individual systems where the threat does not apply. An amber note above the selection names datasets with a bias finding and suggests the threat Bias and discrimination.
Save the scenarios
Save as Draft saves without checks. Back also saves and leads to the chapter entry. Save & Continue checks the scenarios, publishes them, creates a snapshot of the Risk register (AI) as a record and opens the risk analysis. Saving replaces the whole state: a removed scenario takes its rating with it. Scenarios whose group no longer exists after a type change in the inventory appear under Scenarios no longer assigned and can be deleted there with Remove.
Rate the scenarios
Set the scale
At the top of the Risk analysis page sits the Risk scale box with the options Standard AI matrix and ISMS scale. The standard AI matrix rates with one selection each for severity and likelihood on five levels. The ISMS scale adopts the severity dimensions, aggregation rule and likelihood levels from your published ISMS risk policy; the same risk level then means the same thing in both management systems.
Rate for the organisation
The scenarios are grouped by AI type; from two scenarios onwards you can filter by AI type, Threat and State. In the left half of the card, For the organisation, choose the Severity and the Likelihood, on the ISMS scale the severity per severity dimension. On the ISMS scale the likelihood is two-tiered: you estimate the Threat likelihood and set with the slider How likely does the threat succeed? how often an attempt gets through; the field Incident likelihood (calculated) shows the result. Once both axes are set, the card header shows the risk level with the risk score.
Rate for the people affected
In the right half, For the people affected, tick the applicable harms under Type of harm and choose the Severity for the people affected on the same level scale; it appears as the badge Affected: … next to the risk score. Both judgements sit side by side because a scenario can be acceptable for the organisation and severe for the people affected. If a system is linked to a data protection impact assessment, the harms are prefilled from it when the page is first opened. The note Touches vulnerable groups appears as soon as the impact assessment of an involved system lists such a group.
Check the matrix and save
The Risk matrix below the list shows every scenario with both axes set; a gap points to an unrated scenario, which you find with the filter Not yet assessed. Save & Continue publishes the ratings, creates a snapshot of the Risk register (AI) and opens the classification of the AI risks. Back saves silently and leads to the risk identification.
Check the risk levels
Walk through the groups from top to bottom
The page Classifying the AI risks distributes the scenarios according to your scale into the groups under AI Risk Scenarios by Risk Level; above them the Risk Matrix shows the same scenarios as dots. Each card carries the values from the analysis, the risk score and the colour label of the calculated level. Check per scenario whether the calculated level fits what you know about the system and the threat. The levels here are your treatment priority; the risk class from the classification stays untouched.
Reclassify a scenario with a reason
Drag a card into another group. The Override Risk Level dialog requires a Reason for Override from five options, such as assessment uncertainty or qualitative acceptance criteria, and a Justification Comment (required); Confirm Override applies the reclassification. The card gets an amber border and a line with Original level and Reclassification reason; in the matrix its dot turns orange. Drag the card back into its calculated group and the override is removed without a prompt.
Complete the classification
Scenarios without a complete analysis are collected in the greyed-out group Not Yet Assessed (incomplete analysis); complete their values in the risk analysis. Save & Continue requires at least one rated scenario, records the calculated and the effective level per scenario with its justification and opens the risk treatment. Back leads to the risk analysis here without saving; secure reclassifications first with Save as Draft.
Set the controls per system
Review the prefilled matrix
The AI risk treatment page shows a matrix: the rows are the ten Controls from ISO/IEC 42001 Annex A, the columns your classified AI systems, with the badges High-risk or Prohibited in the column header. Cenedril has applied every control that qualifies by role and risk class. The two provider controls Management guidance for AI system development and Data for AI systems are locked as Only relevant for providers on systems without a provider role; on a prohibited system every cell is locked with Use prohibited, system is being discontinued. Clicking a control name opens its purpose, standard reference and individual criteria.
Deselect controls and give reasons
Clicking the tick of a cell removes the control for that system; below it the field Select reason appears with reasons such as Low risk, Does not apply or Other reason with your own text. Two cases turn the cell red without blocking the save: deselecting a control required for high-risk systems on a system marked High-risk shows Art. 27 required, and deselecting Information for interested parties on a system with a transparency obligation shows the note that Art. 50 applies to this system. Both deselections later appear in the Statement of Applicability together with your reason.
Use the row tools
Next to each control name sit two buttons. Apply to all systems sets or removes the control in one go for every open cell of the row. Adjust individual criteria opens the individual criteria of the control; deselected criteria apply to all systems together, and applied cells then appear hatched with the note Applied, some criteria deselected.
Save the treatment
The footer offers Back (to the classification without saving), Cancel (to the wizards overview without saving), Save as Draft and Save & Continue. The last one checks that every deselected open cell carries a reason, publishes the matrix and opens the treatment effects. Publishing creates the records Statement of Applicability (AI) and Risk Treatment Plan under Documentation.
Rate the residual risk
Review the mitigating controls
The page Treatment Effects and Risk Reassessment shows per scenario, in the grey box Mitigating controls, the controls that counter the threat and are at the same time applied to the affected system. The line Stages under Art. 9(5): counts them per stage: Safety by design, Technical protective measure, Information and instruction. If the box says that no mitigating controls have been applied yet, nothing is currently acting on this scenario; that is changed on the risk treatment page.
Record the entry for Article 9(5)(a)
Under Elimination or reduction through design and development (Art. 9(5)(a)) choose whether the hazard was eliminated or reduced in the system itself, whether that was not technically feasible, or Other with a sentence. If a mitigating control carries the stage safety by design, the reduction is prefilled and marked as derived; Clear entry withdraws a choice of your own. If the entry is missing, a note with a jump list appears above the footer; publishing remains possible.
Enter residual values and give reasons
Enter the Residual severity and the Residual likelihood; the original value is shown next to each field, per dimension and with Threat likelihood and slider on the ISMS scale. The field Residual severity for the people affected appears when the analysis recorded a severity. The Risk score updates immediately, as do the card Risk matrix before and after and the Summary at the top. Under Reasoning for the reassessment record which control lowers which factor; this sentence goes into the Risk Treatment Plan. One level per factor is the usual reduction for a control that actually runs.
Decide on acceptance and have it confirmed
Once a residual value is set, the switch Residual risk is within the acceptable range appears. Its default follows the acceptance threshold of your scale, the upper limit of the lowest risk level. If the residual value lies above it, the amber block Residual risk above the acceptance threshold opens with a reason such as Carried for a limited time, further measures are planned and the Explanation of who decided and what is planned by when. The checkbox at the end of the card, with which the responsible person confirms the residual risk, appears only on scenarios with an assigned person; who ticked it and when is recorded.
Set up the annual risk cycle and save
In the block Annual risk cycle, Set reminder (yearly) creates a recurring reminder for the whole cycle; the recipients are prefilled with the people in the AI role AIMS Operation. The reminder is created immediately, independently of saving the page. Save & Continue completes the step without forcing a residual value on every card and opens the impact assessment; this creates the Risk Treatment Plan and a snapshot of the Risk register (AI) as records. Back leads to the risk treatment without saving.
Complete the impact assessment
Review the cards per system
The page Fundamental rights impact assessment lists under Assessment per system one card for every classified system with a high risk class and for every system in your own provider role; the reason is shown top right as High risk or Own provider role. The badge Legally required marks a system where the classification triggered the fundamental rights impact assessment under Article 27 as a legal duty; such cards come first. Two grey sections are derived: How the system is used from the AI register, and From the risk analysis with harms, foreseeable misuse and the highest severity of this system’s scenarios.
Set the basis of the assessment
In the block Basis of the assessment enter First use on and choose under Builds on whether the assessment is Carried out independently, builds on the Impact assessment of the provider (with a reference) or on an assessment for one of your own systems with shared purposes of use; in the last case the card takes over its entries into still empty fields and reports what was taken over. Under Data protection impact assessment (Article 27(4)) you see what the data protection module records for this system, and you connect further activities via Connect a processing activity.
Assess affected groups, consequences and benefits
Under Who can be affected by the use? the catalogue is prefilled from the purposes of use of the system; Add another group of people adds what the catalogue does not know. For each selected group a block appears under Consequences and benefits per group with Consequence areas touched and Benefit for this group. Groups with the badge Requires particular protection then appear on every risk scenario of this system. A bias finding on a linked dataset suggests the impact area fairness.
Record circumstances and societal fields
The question Under which known or foreseeable circumstances can this system, when used as intended, lead to risks to health, safety or fundamental rights? aims at compliant use under aggravating conditions; Other opens a text field. Answer the question Does the use of the system affect societal fields as a whole? with Yes or No; only with Yes does the catalogue of societal fields appear. No is a valid answer and clears the selection.
Close: notification and sign-off
In the block Closing the impact assessment, a card marked Legally required shows the section Notification of the market surveillance authority (Article 27(3)). The recipient is the authority from the Incident settings of the AI Compliance Assistant; the Status of the notification ranges from Not notified yet via Notified with date and file reference to the two cases in which no notification is required. On every card choose under Carried out and owned by the person who is professionally accountable for the assessment and set Completed on. The sign-off covers every entry on the card, including those deliberately left empty.
Set the repetition and complete
The block Repeating the impact assessment below the cards applies to all systems: under How often is the impact assessment repeated? choose between Yearly, Every two years and Only when an occasion arises; the cadence also appears in the impact assessment chapter of the AI policy. The reminder Repeat the impact assessment follows the saved cadence: Set reminder (yearly) or Set reminder (every 2 years) creates it, and with Only when an occasion arises there is none; recipients are the people in the AI role Risk & Impact Assessment. Save & Continue writes a versioned record per card, Fundamental Rights Impact Assessment or Impact assessment with the system name, and opens the chapter Rules, competence & implementation. If writing a record fails, the publish is aborted and your entries are kept as a draft.
Result: every scenario of the chapter is rated, classified and carries a residual value. The Risk register (AI), the Statement of Applicability (AI), the Risk Treatment Plan and the impact assessment per high-risk system or own provider system are stored as records under Documentation in Cenedril, and the chapter row Risks & Impact Assessment in the overview shows the number of scenarios and impact assessments and the distribution of the risk levels.