When an AI tool fails, the response should protect the client file first
Pause affected work, identify the tool, data and downstream reliance, then assign an owner to assess corrections, notification duties and the control gap. Keep an incident record proportionate to the impact.
Use the incident record to explain what happened to a client, regulator, insurer or internal file reviewer. Keep uncertainties visible and pause affected work while the facts are checked.
The safest answer is containment, review and recorded remediation
Pause the affected output, identify the work and data involved, assign a professional owner and record the review. Keep the output out of sending, filing, billing and further use while a competent reviewer checks the facts, risks and client impact.
For low-risk internal work, consider whether correcting the output, logging the event and updating guidance addresses the issue. For client advice, regulated submissions, financial analysis, evidence summaries or confidential documents, examine data exposure, affected decisions, approval of corrections and possible notification duties. Check which duties apply to the incident rather than treating this article as a notification rule.
Review your AI risk controls before incidents repeat
Separate tool failure from unauthorised use
Distinguish a tool failure from unauthorised use. In the first case, an approved or tolerated system produced something unsafe, misleading or unreliable. In the second, someone used a tool, account, browser extension, embedded feature or public AI service outside the firm's policy. Both can happen in the same incident.
For tool failure, examine quality assurance, prompts, data sources, model settings, retrieval logic, human review and escalation thresholds. For unauthorised use, examine access control, training, procurement, approved-tool lists, browser controls, supervision and incentives. Ask staff whether the approved route was clear, timely and suitable for the task; do not assume why they bypassed it.
For either route, record what happened, the affected client or internal work, the data involved, who reviewed the output, what changed and any wider control gap.
Use a practical incident triage checklist
Use these six questions to organise the initial review:
- Scope: Which client, matter, audit, valuation, advice note, report or internal workflow was affected?
- Data: Was confidential, personal, privileged, commercially sensitive or regulated data entered into the tool?
- Reliance: Did anyone rely on the output for advice, analysis, client communication, pricing, evidence, filing or decision support?
- Review: Who is the named professional owner, and what independent check has been completed?
- Disclosure: Does the client, insurer, regulator or internal risk committee need to be informed?
- Prevention: What control changed so the same failure is less likely next time?
Use the answers to distinguish a drafting issue from possible data protection, confidentiality, professional indemnity or client-outcome risks. The checklist supports triage; it does not determine the firm's professional duties.
Build the governance route before the first serious incident
Define approved and prohibited AI uses, review standards, disclosure triggers and escalation routes before an urgent incident. Include a clear route for staff who are unsure whether a use is allowed.
Give teams a clear list of approved tools and uses, data rules, reviewer expectations and out-of-scope work. Consider unusual browser extensions, unexplained document metadata, unapproved accounts, staff feedback and recurring quality issues when deciding what to investigate.
Create a governance route for AI incidents
What the incident record should contain
Include the date, tool, workflow, client or internal area, data category, nature of the failure, reviewer, decision, remediation, disclosure assessment and control change. Keep the record usable for the team and detailed enough to explain the review.
For regulated or high-risk work, assess whether to keep the source output and corrected version under the applicable confidentiality and retention rules. Record where evidence is held and who can access it. Make the reporting route clear so staff know how to raise an issue.
Conclusion: make failure safe to report and hard to repeat
An incident can expose gaps in scope rules, human review, data boundaries or evidence trails. Examine how those gaps affected the work and whether an unsafe output reached a client file.
Assign a named owner, a review standard and a record for AI-assisted client work. When a tool fails or is used without authorisation, preserve the facts, assess client impact and document the control changes agreed after review.
Turn AI governance rules into working implementation steps
Need a safer AI route for your firm?
If this article reflects a live decision in your practice, the useful next step is to map the workflows, confidentiality risks, supplier controls and governance gaps before tools spread informally.
