Purpose
The AI Incident Response Playbook defines how legal departments detect, classify, escalate, and resolve AI-related incidents. It operationalises incident-handling obligations from the AI Use Policy (GOV-02) and the AI Risk Register (GOV-03), which route Level 4 events (materialised risks) into this process. Resolved incidents feed back into GOV-03 risk scores and trigger Defensibility Posture Statement (DPS) updates under GOV-01.
When to use this Module
Activate this playbook immediately upon any of the following:
- Detection of a potential or confirmed AI-related incident
- Receipt of a Shadow AI report under GOV-02 Section 3 protocols
- Level 4 escalation trigger from GOV-03
- Client complaint or external inquiry related to AI use
- Regulatory inquiry or notification obligation arising from AI system behaviour
Owner: Incident Commander (General Counsel or designated deputy)
Duration: Per incident; 15 minutes to activate; resolution timeline by severity
Operating cadence: Continuous (activate on incident; test quarterly)
Scope
This playbook covers all AI incidents involving AI systems listed in the AI Bill of Materials (AI BoM). An AI incident is any event in which an AI system:
- Produces outputs that cause or threaten client harm
- Processes client data in an unauthorised or non-compliant way
- Fails to perform as expected in a matter-critical context
- Is used in violation of the AI Use Policy (Shadow AI)
- Triggers a regulatory or professional responsibility obligation
Section 1 — Incident taxonomy (Risk Taxonomy 2026)
AI incidents are classified against the Risk Taxonomy 2026. Each incident is assigned a primary risk class, which drives escalation, notification, and remediation.
| Risk class | Incident type | Typical trigger |
|—|—|—|
| Class 1: Hallucination and accuracy | AI generates fabricated citations, incorrect legal interpretations, or erroneous contract terms | Attorney discovers AI output error before or after filing / delivery |
| Class 2: Privilege and confidentiality | Client data uploaded to unapproved platform; vendor uses client data for training; privileged material accessed without authorisation | Data audit; client complaint; vendor notification |
| Class 3: Bias and fairness | AI produces discriminatory outputs in legal research, contract analysis, or advice | QA review; client feedback; internal bias audit |
| Class 4: Privacy and data protection | AI system processes personal data in breach of GDPR, CCPA, or other applicable law | DPIA review; regulatory enquiry; data subject complaint |
| Class 5: Supply chain and vendor dependency | Vendor breach, service outage, model change, or compliance failure affecting legal work | Vendor notification; service monitoring alert; contract review |
| Class 6: Shadow AI and policy circumvention | Personnel use unapproved AI tools or circumvent AI Use Policy controls | Network monitoring alert; anonymous report; GOV-02 attestation gap |
| Class 7: Regulatory compliance drift | New obligation (EU AI Act, ABA ethics opinion, state law) creates non-compliance with current practice | Regulatory publication; bar association bulletin; regulatory inquiry |
| Class 8: IP and licensing exposure | AI-generated work product creates IP infringement risk or lacks copyright protection | Counsel review; client question; vendor licensing change |
| Class 9: Operational resilience | AI system outage, degraded performance, or agentic AI unexpected action affecting matter work | System monitoring alert; user report; agentic AI log anomaly |
Section 2 — Severity levels and response windows
Level 1 — Critical (immediate response)
Definition: Confirmed or highly probable exposure of privileged client data; professional responsibility violation with external consequence; AI error in a filed or delivered document; regulatory investigation initiated; agentic AI system takes unauthorised external action.
Response windows:
- Initial containment: Within 1 hour of detection
- Incident Commander notified: Immediately
- Full response team assembled: Within 2 hours
- Executive briefing: Within 4 hours
- Client notification assessment completed: Within 6 hours
- Regulatory notification: Per applicable law (typically 24–72 hours)
Mandatory escalations: General Counsel (immediate); CEO/Executive Leadership (within 4 hours); Board (within 24 hours); Professional Liability Insurance Carrier (within 24 hours); External Counsel (as required).
Level 2 — High (urgent response)
Definition: Significant AI bias discovered in delivered work product; Shadow AI use involving client data; vendor security incident affecting client data; AI accuracy failure on an important matter; confirmed compliance violation.
Response windows:
- Initial response: Within 4 hours
- Incident Commander notified: Within 2 hours
- Response team assembled: Within 8 hours
- Legal counsel briefed: Within 6 hours
- Client notification assessment: Within 12 hours
Mandatory escalations: General Counsel (within 4 hours); Legal Operations leadership (within 2 hours); IT Security Manager (within 2 hours); Practice group lead for affected matters (within 8 hours).
Level 3 — Moderate (standard response)
Definition: AI tool behaves unexpectedly but no client data exposed; Shadow AI use without client data; vendor compliance issue without immediate client impact; regulatory change requiring policy update.
Response windows:
- Initial response: Within 24 hours
- Incident Commander notified: Within 8 hours
- Response team assembled: Within 48 hours
Mandatory escalations: AI Task Force (within 24 hours); Legal Operations (within 48 hours).
Level 4 — Minor (standard review)
Definition: Training gap identified; near-miss with no client impact; policy clarification request; AI tool performance degradation within acceptable thresholds.
Response windows:
- Document and assign: Within 72 hours
- Review at next AI Risk Committee meeting
Escalations: AI Task Force log; no mandatory escalation.
Section 3 — Response team
| Role | Default holder | Primary responsibilities |
|—|—|—|
| Incident Commander | General Counsel (or designated deputy) | Activates playbook; authorises resource deployment; owns external communications |
| Legal Lead | Responsible attorney or Practice Group Head | Professional responsibility assessment; privilege protection; client notification strategy; litigation hold decisions |
| Technical Lead | IT Security Manager | Technical investigation; system containment; forensic evidence collection; vendor coordination; system restoration |
| Privacy Officer | Chief Privacy Officer or Compliance Officer | GDPR/CCPA compliance; regulatory notification; data subject rights; DPA coordination |
| Communications Lead | Legal Operations or designated comms contact | Internal communications; client communication support; regulatory correspondence support |
| AI Task Force Liaison | AI Task Force representative | GOV-03 risk register update; GOV-02 policy implications; AI BoM status; DPS update trigger |
Level 1 incidents require full team activation. Level 2 requires Incident Commander, Legal Lead, and Technical Lead at minimum. Level 3 requires Incident Commander and Technical Lead. Level 4 requires AI Task Force Liaison only.
Section 4 — Activation sequence
Step 1 — Detect and classify (0–15 minutes)
- Person detecting the incident notifies the AI Task Force via the designated incident reporting channel.
- AI Task Force assigns an incident risk class (Class 1–9) and initial severity level (Level 1–4).
- Incident Commander is notified per the response window for that severity level.
- AI BoM is consulted to identify the specific AI system involved.
Step 2 — Contain (15 minutes – 2 hours)
- Technical Lead executes immediate containment: isolate affected system; revoke access as required; preserve logs.
- Legal Lead assesses privilege implications and initiates litigation hold if applicable.
- Privacy Officer assesses data protection notification obligations.
- AI BoM record for the affected system is flagged as under investigation.
Step 3 — Assess and notify (2–24 hours)
- Incident Commander conducts a formal severity assessment with the full response team.
- Client impact assessment: identify affected matters and clients; assess harm.
- Notification matrix applied: clients, regulators, bar associations, insurer as required by law and severity.
- GOV-03 risk register entry for the relevant risk class is updated with incident data.
Step 4 — Remediate (24 hours – resolution)
- Technical Lead implements root cause remediation.
- Legal Lead oversees client remediation and privilege repair where possible.
- Privacy Officer oversees regulatory remediation and documentation.
- Affected AI system remains suspended from the AI BoM until remediation is certified.
Step 5 — Close and review
An incident is closed when:
- Root cause is identified and documented.
- Remediation actions are completed and verified.
- All notification obligations are fulfilled and documented.
- GOV-03 risk register is updated with revised risk score for the relevant class.
- AI BoM status for the affected system is resolved (reinstated, modified, or removed).
Section 5 — Post-incident review and DPS update
Post-Incident Review (PIR)
A Post-Incident Review is conducted within 30 days of closure for Level 1 and Level 2 incidents. The PIR produces:
- Root cause analysis and timeline.
- Control gaps identified in GOV-02 (AI Use Policy) or GOV-03 (Risk Register).
- Policy or control updates required.
- Training needs identified.
- DPS update assessment.