AI Council Toolkit
Operations

Incidents

Managing AI-specific incidents and near-misses.

What Counts as an AI Incident?

An AI incident is any event where an AI system causes or contributes to harm, near-harm, or a significant deviation from expected behavior. Examples:

  • A model produces biased or discriminatory outputs in production
  • An AI system makes a consequential error affecting a real person
  • A data breach involves training data or model outputs
  • A generative AI system produces harmful, illegal, or reputation-damaging content
  • A system fails in a way that disrupts business operations
  • A near-miss is identified before harm occurs

Incident Response Process

1. Report

Anyone should be able to report an AI incident or near-miss. Channels should include:

  • A dedicated email or form
  • The champion network (first point of contact in most teams)
  • Existing incident management tools (e.g., ServiceNow, PagerDuty)

2. Triage

The council chair or on-call designee triages the incident:

SeverityDescriptionResponse Time
CriticalActive harm to individuals, legal exposure, public-facingImmediate
HighSignificant risk of harm, regulatory implicationsWithin 24 hours
MediumDegraded performance, internal impactWithin 3 business days
LowMinor anomaly, no direct harmNext scheduled review

3. Contain

  • Stop the system from causing further harm (pause, roll back, add human review)
  • Notify affected individuals if required
  • Preserve evidence (logs, inputs, outputs)

4. Investigate

  • What happened?
  • What was the root cause?
  • Was this foreseeable? Was it in the risk assessment?
  • What controls failed or were missing?

5. Remediate

  • Fix the immediate issue
  • Update controls, monitoring, and risk assessment
  • Implement preventive measures

6. Review and Learn

  • Present findings at the next council meeting
  • Update the incident log
  • Update policies, templates, or training if needed
  • Share anonymized lessons learned with the champion network

Incident Log Template

FieldDescription
Incident IDUnique identifier
Date reportedWhen the incident was reported
SystemAI system name and ID
SeverityCritical / High / Medium / Low
DescriptionWhat happened
ImpactWho was affected and how
Root causeWhat went wrong
Actions takenContainment and remediation steps
Lessons learnedWhat the organization should do differently
StatusOpen / Investigating / Resolved / Closed

Copy This Template

# AI Incident Report

**Incident ID:** [INC-NNN]
**Date reported:** [Date]
**Reported by:** [Name and role]

## System Information

| Field | Response |
|-------|----------|
| **System name** | |
| **System ID** | |
| **System owner** | |

## Incident Details

| Field | Response |
|-------|----------|
| **Severity** | [ ] Critical [ ] High [ ] Medium [ ] Low |
| **Date/time of incident** | |
| **Date/time detected** | |
| **Description** | |
| **Impact** | |

## Response

| Field | Response |
|-------|----------|
| **Containment actions** | |
| **Individuals notified** | |
| **Evidence preserved** | |

## Investigation

| Field | Response |
|-------|----------|
| **Root cause** | |
| **Was this foreseeable?** | |
| **Was it in the risk assessment?** | |
| **What controls failed?** | |

## Remediation

| Field | Response |
|-------|----------|
| **Immediate fix** | |
| **Preventive measures** | |
| **Policy/template updates needed** | |
| **Lessons learned** | |

**Status:** [ ] Open [ ] Investigating [ ] Resolved [ ] Closed

On this page