AI Risk Register - Tracking and Mitigating LLM Risks

ShopMax India's LLM systems handle product recommendations, customer support, pricing queries, and fraud detection. Each use case carries distinct risks - hallucinated product details, biased recommendations, data leakage, or regulatory non-compliance. An AI risk register is a structured inventory of these risks, rated by likelihood and impact, with assigned owners and mitigation actions. It is the cornerstone of a responsible AI governance programme.

Each risk entry captures the risk description, affected system, likelihood score (1-5), impact score (1-5), risk score (likelihood x impact), mitigation strategy, owner, and review date. The register is reviewed monthly. High-score risks trigger immediate mitigation sprints. ShopMax India integrates the risk register with its incident response system so production incidents automatically create or escalate risk entries.

The example below builds a lightweight AI risk register for ShopMax India using Python dataclasses with automatic priority scoring and CSV export for stakeholder reporting.

It gives the following output,

ID       System                 Score   Priority   Owner
--------------------------------------------------------------
R-001    Product Recommender    12      HIGH       ML Team
R-004    RAG Pipeline           12      HIGH       Infra Team
R-002    Support Chatbot        10      HIGH       Security Team
R-003    Pricing Bot            6       MEDIUM     Data Team

Exported to ai_risk_register.csv

Review the risk register monthly with representatives from ML, security, legal, and product teams. Any production incident should trigger a risk register review to check whether the incident was a known risk that escalated or a new risk to add. Set automated alerts when a risk score rises above 15 - the owning team must submit a mitigation plan within 48 hours. Archive old risk entries rather than deleting them to maintain a complete audit trail for regulators.