Respond to the system, not only the model
An AI incident can cross the model provider, application, retrieval store, user identity and connected business system. Atlant Security helps scope that chain, coordinate authorised containment and establish evidence for recovery decisions.
Choose the response workstream
- Prompt Injection Incident Response — Investigate malicious instructions reaching an AI application. Trace retrieved content, tool calls and downstream effects before defining containment.
- AI Agent Incident Response | Tools, Identities & Actions — Respond to agents making unauthorised tool calls, changing records or accessing unexpected systems. Establish the action trail and contain delegated access.
- AI Data Leak Response | Scope Exposure & Preserve Evidence — Investigate sensitive information exposed through AI outputs, retrieval or connected tools. Establish affected data, recipients and evidence limitations.
- RAG Poisoning Response | Isolate Untrusted Retrieval Sources — Investigate poisoned documents, manipulated retrieval indexes and persistent agent memory. Trace source provenance, ingestion changes and affected outputs.
- AI API Key Compromise | Revoke Access & Investigate Abuse — Respond to exposed AI API credentials, unexpected usage and abused service accounts. Establish permission scope, usage history and safe credential replacement.
- Microsoft 365 Copilot Incident Response | Access & Data Exposure — Investigate suspected Copilot-related data exposure, malicious content and connector abuse. Review Microsoft 365 identities, permissions and available evidence.
An engagement with clear boundaries
We agree affected services, named decision makers, access permissions, evidence handling and communication before taking action. Work can cover investigation, containment support, recovery validation and post-incident improvements. Provider access, specialist support and operating hours are confirmed for the specific engagement.
What the enquiry activates
An enquiry asks the team to confirm availability and scope. It does not automatically dispatch a responder, authorise containment or establish a response-time commitment. Existing incident processes and contracted emergency channels remain in effect.
From the first question to a recovery decision
| Stage | Work product | Customer involvement |
|---|---|---|
| Triage and preservation | Agreed incident boundary, evidence priorities and immediate decisions | Incident lead and authorised technical owners |
| Investigation | Correlated timeline, impact assessment and unresolved questions | Access to relevant records and provider escalation |
| Containment and recovery | Approved change record and acceptance evidence | Change approvers and service owner |
| After the incident | Prioritised improvements and revised runbook | Owners for implementation and validation |
