Nonconformance

Purpose

This procedure sets how Rygen identifies, records, investigates and corrects nonconformities in its AI Management System (AIMS).

Scope

This procedure applies to nonconformities in AI systems within the AIMS scope (X1, Corsair and internal AI tools), in AIMS processes and procedures, in AI governance and compliance requirements, and in AI system performance, security or ethics.

Definitions

Nonconformity: Failure to fulfill a requirement of the AIMS. Where a requirement is not met, this includes AI system failures or performance degradation, policy violations or process deviations, compliance gaps or regulatory issues, data governance or security breaches, and stakeholder complaints about AI systems.

Corrective Action: Action taken to eliminate the cause of a nonconformity and prevent recurrence.

Nonconformity Identification and Immediate Response

A nonconformity must be contained and documented as soon as it is identified.

Identification Sources

Nonconformities may be found through AI system monitoring and alerts, internal audits and reviews, stakeholder feedback or complaints, risk assessment, incident reports, and AIMS performance evaluation.

Immediate Response

The person who identifies a nonconformity shall:

  1. Control and correct it. Stop or contain the nonconforming activity if it is safe to do so, put safeguards in place to prevent further impact, and record the actions taken.
  2. Deal with the consequences. Assess and reduce the impact on stakeholders, tell affected parties as required, and put temporary measures in place if needed.
  3. Create a Bug in Jira that describes the nonconformity, and assign it to the relevant team lead or the Principal AI Engineer. Set its priority:
    • Highest: critical AI system failure, security breach or regulatory violation.
    • High: significant impact on AI system performance or stakeholder trust.
    • Medium: moderate impact that needs prompt attention.
    • Low: minor issue with limited impact.
    • Lowest: documentation or process improvement.

Root Cause Analysis and Corrective Action Planning

Once a nonconformity is contained, it must be analyzed and corrected.

Evaluation of Need for Action

The Principal AI Engineer, or a designee, shall decide whether corrective action is needed by:

  1. Reviewing the nonconformity: its nature and scope, the chance it recurs, and any regulatory or compliance implications.
  2. Determining its root causes with Five Whys analysis, including systemic factors, process gaps and contributing factors.
  3. Determining whether similar nonconformities exist or could occur, by searching Jira for related issues and reviewing patterns, trends and systemic issues.

Root Cause Analysis Methods

Five Whys analysis is required for every root cause analysis. A fishbone diagram, for complex issues with several factors, and timeline analysis, for incident-based nonconformities, may be used alongside it.

Implementation and Follow-up

Corrective actions must be implemented, then reviewed to confirm they removed the root cause.

Implement Corrective Actions

Implement the actions the analysis calls for. These may be process changes, training, system changes, policy or procedure updates, or added controls or monitoring.

Review Effectiveness

Review whether the corrective actions worked, through follow-up audits or assessments, performance monitoring, stakeholder feedback, or trends in similar issues.

AIMS Updates

Make any changes to the AIMS that the corrective action needs, such as updates to the risk register, procedures, training material or policies.

Documentation Requirements

Non-Conformance Report (Jira)

Every nonconformity shall be tracked in Jira with its description, priority, root cause analysis results, planned and implemented corrective actions, effectiveness review results, status updates and closure.

Corrective Action Report

A High or Highest priority nonconformity also gets a Corrective Action Report (CAR-xxx) in the aims-governance repository. The CAR is the single full record of the issue, its investigation and its resolution. It contains:

  • The nonconformity: its source, ISO clause, severity, affected systems and the procedure concerned.
  • The root cause analysis: the method used, contributing factors, whether similar nonconformities exist or could occur, and the blameless root cause.
  • The containment, remediation and systemic improvement actions taken, with their Jira issues.
  • The verification of effectiveness: criteria, method and results.
  • Supporting evidence, including any AI-023 postmortem.

Roles and Responsibilities

All Personnel

Report nonconformities promptly, take immediate corrective action within their authority, and cooperate with investigations and root cause analysis.

Team Leads

Make sure their team knows how to report nonconformities, support immediate response and containment, and take part in root cause analysis for their area.

Principal AI Engineer

Oversees evaluation and response, approves corrective action plans, makes sure the records are complete, and reports significant nonconformities to the CTO and the AI Governance Committee.

AI Governance Committee

Reviews the major AIMS changes that result from corrective actions. Each such change is approved by the authority AI-006, AI-008 or AI-013 names.

Integration with AIMS Processes

Risk Management

Update risk assessments, add new risks and change risk treatments as the findings require.

Monitoring and Measurement

AIMS performance reporting includes nonconformity metrics, corrective action effectiveness and trends.

Management Review

Nonconformity trends and corrective actions are reported to the CTO in each management review report (AI-012), and inform AIMS improvement priorities.

Records and Evidence

Required Documentation

Keep documented evidence of:

  • The nature of each nonconformity and the actions taken: Jira issues and their updates, and any AI-023 postmortem.
  • The results of corrective actions: effectiveness reviews in Jira and follow-up verification records.

Record Retention

  • Jira issues: 3 years after closure.
  • Corrective Action Reports: permanently, in the aims-governance repository.
  • Supporting documentation: at least 3 years.

Revision History

VersionDateAuthorSummary of Change
1.02025-06-05Field BradleyInitial draft.
1.12025-09-02Field BradleyMigrated to markdown and gitlab
1.22026-01-14Field BradleyConsolidated NCR/CAR documentation into single CAR document in aims-governance repository
1.32026-07-14Field BradleyGovernance committee review cadence changed from quarterly to every four months (AI-1759)
1.42026-11-04Field BradleyNonconformity trends reported in the management review report, as the AI Governance Committee holds no regular reviews (AI-2173)
2.02026-11-09Field BradleyShortened every section, keeps Five Whys as the required method, matches the CAR contents to the template, and cites AI-006 and AI-012 for approvals and schedule (AI-2173, AI-2131). Removes the AI Governance Committee’s review of High and Highest priority nonconformities, and its monitoring of trends, which management review covers. Evidence includes any AI-023 postmortem.