In case of an AI incident, is your organisation prepared?
One warning. Three decisions. Take command.
Autonomous agent Critical
Your AI agent initiates unauthorised payments.
An alert flags payments to a new supplier. Your agent initiated them without human approval and can still initiate more.
Your role: Operations director
Contain
A tested stop mechanism, restricted access and preserved logs.
Coordinate
An incident owner and an escalation path across finance, security, leadership and relevant partners.
Recover
Isolated testing, transaction limits, human approval and monitored alerts.
Data exposure High
A client file was shared with an unapproved AI tool.
A team member uploaded a file containing client information to a public AI tool. Its retention terms are unknown.
Your role: Operations lead
Contain
A reporting procedure, an approved-tool inventory and preserved evidence for analysis.
Coordinate
A documented assessment with privacy, security and legal owners, including applicable obligations.
Prevent recurrence
Approved tools, data minimisation rules and training matched to actual use.
Recruiting bias Significant
Your recruiting AI ranks comparable candidates differently.
Before approving a shortlist, your team notices unexplained differences between candidates with comparable qualifications.
Your role: HR director
Protect decisions
A way to suspend recommendations and human review against job-related criteria.
Understand
Documented criteria, tests suited to affected populations and review with HR, technical and legal owners.
Control the restart
Acceptance criteria, explicit human responsibility and regular review of decisions.
Customer chatbot Moderate
Your chatbot promises delivery dates you cannot meet.
Several customers received delivery dates invented by your chatbot. The team discovers the discrepancies when customers call.
Your role: Customer service lead
Correct
Human handoff and the ability to disable unreliable answers on a specific topic.
Restore trust
A correction owner, verified information and a process for communicating with affected people.
Make reliable
Current sources, test cases, refusal to answer without reliable data and a reporting procedure.
Internal summary Low
An AI summary misses an important action.
An AI-generated internal meeting summary omits an action and its owner. It is still a draft and has not been shared.
Your role: Team lead
Verify
A check of actions, owners and deadlines against the source notes.
Clarify
Action validation by the people involved and a contact for correcting errors.
Improve practice
A short pre-sharing verification routine and regular review of observed errors.
- AI system and agent evaluation Test system limits, document risks and prioritise controls.
- AI governance and risk management Clarify responsibilities, decision rules and escalation procedures.
- Responsible AI Practice training Prepare teams to verify, report and act in their everyday AI use.
- AI practice support and oversight Review controls and practices as your AI use evolves.