AI Risk Treatment Plan
Purpose: Define how AI risks are treated through phased visibility, controls, runtime protection, agent governance and continuous compliance.
Audience: Risk owners, governance, security engineering, architecture review boards.
Treatment Strategy
| Phase | Objective | Key Activities | Outcome |
|---|---|---|---|
| Phase 1: Visibility and Discovery | Understand AI usage and AI assets | Discover AI tools, users, endpoints, IDE plugins, SaaS AI, AI APIs and custom AI apps | AI inventory and usage baseline |
| Phase 2: Data Protection and Enforcement | Reduce leakage through AI interactions | Enforce classification-aware prompt and upload controls; detect source code and credentials | Data leakage risk reduction |
| Phase 3: Runtime Protection | Secure custom AI applications | Test prompt injection, RAG retrieval, output governance, AI API security | Production-ready AI application controls |
| Phase 4: Agentic AI Governance | Prevent unauthorized AI-driven actions | Define agent identity, tool/API permissions, approval workflows, action logs | Controlled agentic AI adoption |
| Phase 5: Continuous Governance | Maintain evidence and oversight | Periodic review, dashboard reporting, exception tracking, audit evidence | Sustainable AI governance maturity |
Risk Treatment Options
| Option | When to Use | Required Evidence |
|---|---|---|
| Mitigate | Control can reduce likelihood or impact | Control configuration, test results, logs |
| Avoid | Risk exceeds tolerance or controls unavailable | Rejection decision, architectural notes |
| Transfer | Third party assumes contractual responsibility | Contract terms, SLA, audit rights |
| Accept | Residual risk is within tolerance | Risk acceptance record, owner approval, review date |
Exception Handling
All exceptions must include business justification, affected AI platform, data classification, compensating controls, approval owner, expiry date and review cadence.