Name the incident owner
Identify who can coordinate the application, identity, data and infrastructure owners. Establish the affected service and whether there is continuing operational harm. Use existing incident channels and supplier contacts while response support is being arranged.
Record observation separately from interpretation
Write down what was seen, when it was seen and which record supports it. “The assistant displayed confidential text” is different from “an external recipient received a file”. Preserve the original observation and assign hypotheses for investigation.
Preserve volatile context deliberately
Relevant evidence can include a workflow version, run identifier, retrieved document, tool response and destination event. Ask authorised owners about retention and exports. Collect only what the investigation needs through agreed channels; screenshots alone may omit timestamps, identity and action context.
Choose a containment boundary
Consider the affected capability rather than automatically disabling every AI service. A connector, credential, workflow or ingestion route may provide a practical boundary. The owner should weigh continuing harm, business continuity and evidence preservation before approving changes.
Set the next decision
Define the immediate question: is activity ongoing, which access is implicated, or what is needed for a safe pause? Record uncertainty and responsibilities. The public brief builder can organise a request, but it cannot establish compromise or activate response coverage.
A first-update structure
| Field | Useful entry |
|---|---|
| Observation | What was seen, by whom and at what time |
| Impact | Affected service and confirmed consequence |
| Status | Ongoing, paused or not yet established |
| Evidence | Record reference and collection owner |
| Decision | Action needed, approver and next update |
Use our AI response plan to turn these questions into an agreed scope.
Prepare the next step
Explore AI incident response from Atlant Security, use the evidence checklist or prepare a non-sensitive incident brief for the team.
Sources and further reading
- OWASP GenAI Incident Response Guide
- Microsoft AI agent shared responsibility
- Microsoft prompt injection guidance
- NIST incident response recommendations, SP 800-61 Rev. 3
Reviewed 23 October 2026. These sources inform the approach; they do not certify or endorse Atlant Security.
