Risk Management

Purpose

This document establishes the framework for identifying, assessing, evaluating, and treating risks within Rygen Technologies’ AI Management System (AIMS), in accordance with ISO 42001, which ensures that AI risks are addressed systematically, ensuring responsible AI delivery that our stakeholders can depend on.

Scope

This framework applies to all AI systems within the AIMS scope, personnel involved in the AI system development lifecycle, and all risk assessments (initial, periodic, or trigger-based).

Definitions

Risk: The potential for an uncertain event or condition to have a positive or negative effect on AI system objectives, stakeholder interests, or organizational goals.

Likelihood: The probability that a risk event will occur, assessed on a scale from Very Low (1) to Very High (5).

Impact: The magnitude of consequences if a risk materializes, considering financial, operational, reputational, compliance, and stakeholder effects.

Risk Level: The combination of likelihood and impact (Likelihood × Impact), determining the priority and required response for each risk.

Risk Treatment: Actions taken to modify risk through acceptance, mitigation, avoidance, or transfer strategies.

Residual Risk: The risk level expected once the treatment plan is implemented; the treatment status in the Risk Treatment Register shows which treatments are in place.

Risk Methodology

We take a systematic and phased approach to managing risk. This methodology ensures that all risks are properly assessed, weighed against our tolerance, and controlled as necessary to ensure that the risks do not negatively impact Rygen, our clients, or any other interested parties.

Risk Formula & Scales

Risk is calculated using the formula: Risk Level = Likelihood × Impact.

Likelihood Scale

LevelScoreProbability
Very Low1<5%
Low25-25%
Medium325-50%
High450-75%
Very High5>75%

Factors to consider when assessing likelihood include historical data, control effectiveness, and environmental factors.

Impact Scale

LevelScoreExample
Very Low1Minor process adjustments; <$10K financial impact.
Low2Single feature affected; $10K-$50K financial impact.
Medium3Multiple features/clients affected; $50K-$200K financial impact.
High4Major service disruption; $200K-$500K financial impact.
Very High5Business model threatened; >$500K financial impact.

Impact assessments must consider financial, operational, individual (privacy, safety, fairness), reputational, and compliance consequences.

Risk Levels & Required Action

Risk LevelScore RangeTreatment Required
Very Low1-2Monitor only.
Low3-4Document basic controls.
Medium5-9Implement treatment plan.
High10-16Immediate treatment required.
Critical20-25Stop activity until treated.

Risk Status

Each risk is assigned a status to indicate its lifecycle stage:

  • Potential: May occur if a specific event is triggered.
  • Emerging: Is likely to materialize soon and requires planning.
  • Current: Has been realized and must be actively managed.
  • Retired: No longer managed, because its system or capability was retired, it is covered by another risk, or it was found not to be an AI risk. The record is kept and states the date and reason it was retired.

Risk Appetite and Tolerance

Rygen Technologies has a moderate risk appetite, balancing innovation with prudent risk management. Our tolerance is very low for compliance risks.

Risk Tolerance & Authority

Risk LevelRequired ActionApproval Authority
Very Low / LowMonitor or document controls.Principal AI Engineer
MediumImplement treatment plan.Principal AI Engineer
HighImmediate treatment required.CTO
CriticalStop activity until treated.CEO and CTO

The approval authority applies at two points:

  • Treatment: a risk’s treatment plan, or its acceptance, is approved by the authority for the risk’s level before treatment. The CTO approves the treatment plan of any risk rated High before treatment, and the CEO and CTO approve that of any risk rated Critical.
  • System approval: a system is approved by the authority for the highest residual risk level in its risk assessment, and that approval accepts the system’s residual risk. AI-013 (Decision Authority) adds the other approvals a deployment needs.

Delegation of Approval Authority

The CTO accepts High level AI risks and approves their treatment plans, and gives that approval in the merge request; this authority is not delegated. The Principal AI Engineer carries out the detailed review and recommendation that the approval relies on. Critical risks are accepted by the CEO and CTO together, and this authority is not delegated.

An approval by the CTO on a risk assessment signifies that they have been briefed on the key findings and are formally accepting the residual risk based on the recommendation and due diligence performed by the Principal AI Engineer. Each risk report is approved by the authority that Risk Tolerance & Authority sets for its level before treatment. The Principal AI Engineer and relevant Team Leads implement the controls.

Risk Assessment and Treatment Process

Our risk management process follows five phases, from preparation to ongoing monitoring.

Phase: Preparation & Identification

Objective: To identify potential risks to the AIMS.

The assessment lead, guided by the Principal AI Engineer, identifies risks by reviewing organizational context, stakeholder concerns, the AI system lifecycle and the impact assessment, across technical, governance, business and stakeholder categories. Use discussions or workshops where useful.

Impact Assessment Inputs

Where an AI System Impact Assessment (per AI-009) has been completed for the system under review, the negative consequences identified in that assessment must be considered as candidate risks during identification. The severity and stakeholder context documented in the impact assessment should inform the risk analysis. Refer to AI-009 for the full integration and traceability requirements between impact assessments and risk assessments.

Each identified risk is documented in the Risk Register with a unique ID, description, owner, and status. Each register entry is a risk report, which holds the risk’s analysis and treatment plan. The system risk assessment lists the system’s risks and records what became of each impact from the impact assessment; it refers to the risk reports rather than repeating them.

Phase: Analysis & Evaluation

Objective: To determine the level of each risk and decide on a course of action.

For each risk, the owner assesses Likelihood and Impact using the scales in the Risk Methodology section. The Inherent Risk Level is calculated, and existing controls from the ISO 42001 standard are documented.

Based on the risk level and our risk appetite, a treatment strategy is chosen:

  • Accept: The risk is within tolerance.
  • Mitigate: The risk must be reduced.
  • Avoid: The activity causing the risk will be stopped.
  • Transfer: The risk will be shifted to a third party (e.g., via insurance).

Phase: Risk Treatment

Objective: To reduce risk to an acceptable level.

For risks requiring mitigation, a Risk Treatment Plan is documented within the detailed risk report. This plan outlines:

  • The selected treatment strategy and justification.
  • The applicable Annex A controls from the Statement of Applicability (AI-022).
  • The specific treatment measures required to reduce the risk to an acceptable level.
  • The target residual risk level after treatment.

Treatment plans and the acceptance of residual risk must be approved as the Risk Tolerance & Authority section sets.

Risk Treatment Register

The Risk Treatment Register is maintained in the treatments.yaml file within the AIMS Governance repository, which is the same repository where risk reports are registered. Treatment measures are specific implementations that address identified risks; they are not additional SoA controls. The Statement of Applicability (AI-022) contains only the ISO 42001 Annex A controls.

Each treatment in the register includes:

  • ID: Unique identifier for the treatment (e.g., AC.34).
  • Name: Descriptive name of the treatment measure.
  • Status: Current implementation state of the treatment.
  • Implements: The Annex A control(s) that this treatment implements.
  • Evidence: Documentation demonstrating treatment effectiveness.

Treatment Status

Each treatment is assigned one of three statuses:

  • Planned: Treatment has been identified but is not yet implemented.
  • Active: Treatment is implemented, operational, and actively protecting AI systems.
  • Inactive: Treatment has been deprecated, superseded, or deferred.

Treatments are mapped to ISO 42001 Annex A controls via the implements field and are referenced in risk treatment plans. The status of treatments is updated as part of the risk monitoring and review process to ensure that residual risk assessments remain accurate.

Phase: Documentation & Reporting

Objective: To maintain records and inform stakeholders.

All risk assessment and treatment activities are documented in the Risk Register. The register serves as the central record for risk status, treatment plans, and residual risk levels.

Risk Assessment Verification Checklist

Before finalizing any risk assessment document, the assessment lead must complete the following verification checklist:

  • Control and Treatment Reference Validation: All Annex A control references (e.g., A.10.3) have been verified against the Statement of Applicability (AI-022), and all treatment references (e.g., AC.34) have been cross-checked against the Risk Treatment Register (treatments.yaml) to ensure:
    • The control or treatment ID exists and is correctly referenced
    • The treatment status (Planned/Active/Inactive) is accurately reflected
    • The treatment evidence is current and appropriate for the risk treatment plan
  • Risk Treatment Alignment: The selected controls and treatment measures appropriately address the identified risk and support the target residual risk level
  • Approval Authority: The risk level and treatment plan have been routed to the appropriate approval authority per the Risk Tolerance & Authority matrix
  • Documentation Completeness: All required fields in the risk assessment template have been completed

This verification must be completed before the document is submitted for approval.

Reporting Schedule

Key risk information is reported to leadership according to the following schedule:

  • As they arise: High/Critical risks to the CTO, via Slack and the merge request that seeks approval.
  • At each management review: Risk register review (AI-012).

Phase: Monitoring & Review

Objective: To ensure the framework remains effective and risks are managed over time.

Assessments are performed on a set schedule or when triggered by specific events.

  • Scheduled: At each management review, on the schedule AI-012 sets. Risk register review considers the current risk assessments and available evidence of treatment effectiveness, and identifies changes or decisions needed. The management review report records the result: the register position and each risk whose level, status or treatment changed or that needs a decision. Each risk record holds its current assessment, and its Git history holds every change, so risk records and system risk assessments carry no review dates.
  • Triggers: New AI systems, major changes, security incidents, or new regulatory requirements.

Revision History

VersionDateAuthorSummary of Change
1.02025-06-05Field BradleyInitial draft.
1.12025-09-02Field BradleyMigrated to markdown and gitlab
2.02025-09-25Field BradleyAdded details of authority and delegation by the CTO
2.12025-10-01Field BradleyAdded Controls Register section with status tracking (Planned/Active/Inactive)
2.22025-10-21Field BradleyUpdate schedule to only review risk register quarterly during management review
2.32025-10-28Field BradleyAdded Risk Assessment Verification Checklist requiring cross-check of control references against controls.yml (CAR-002)
2.42026-01-16Field BradleyAdded explicit quarterly review dates aligned with AI-012 management review schedule (CAR-008)
2.52026-02-04Field BradleyRenamed Controls Register to Risk Treatment Register; updated references from controls.yml to treatments.yaml; clarified that treatments implement Annex A controls and are not additional SoA items (AI-1216)
2.62026-07-14Field BradleyAligned risk-register review with the every-four-months management review schedule; review dates changed from Apr/Jul/Oct/Jan to Jan/May/Sep (AI-1759)
2.72026-11-02Field BradleyLow risk approval authority moved from the Team Lead to the Principal AI Engineer, matching AI-013; risk analysis held once, in the risk report; verification result recorded in the merge request review; risk records no longer carry next_review dates, because the management review report records each review (replaces the date alignment from CAR-008); High/Critical risks reported to the CTO as they arise rather than monthly, and the four-monthly risk report given at the management review rather than to the AI Governance Committee, which holds no regular meetings (AI-2173) Added the Retired risk status used by the risk register.
2.82026-11-09Field BradleyCritical risks need the CEO and CTO, High-risk treatment plans the CTO, system approval follows the highest residual, reporting goes to the CTO and each management review, and workshops are optional (AI-2131). Removes treatment timelines that nothing tracks, and states what the register review considers.