With the rapid adoption of Generative AI in our marketing department, I’m seeing new vulnerabilities like prompt injection and data leakage. As a CRISC professional, should I be creating a separate "AI Risk Framework," or is it better to integrate these into our existing IT Risk Assessment processes? I’m worried that treating it as "special" will lead to silos in our GRC reporting.
3 answers
Integration is always the better path for long-term governance. In late 2024, my firm updated our "Technology and Security" domain controls to include AI-specific sub-categories. We didn't build a new silo; we just expanded our existing "Data Privacy" and "Third-Party Risk" assessments to ask questions about model training data and output validation. If you treat AI as a separate entity, you'll likely miss the "interconnected" risks—like how an AI vulnerability could lead to a traditional SQL injection. Stick to the CRISC principle of a unified enterprise risk view.
Do your current "Acceptable Use Policies" cover AI tool usage, or are you waiting for the Risk Assessment to be finished first? Often, the policy change is the easiest "Quick Win" risk response while you're still figuring out the technical controls.
AI is just another "Technology" asset in the CRISC Domain 4. Apply the same lifecycle controls you would for any other critical software or cloud service.
Exactly, Michael. The "Asset Identification" step of the risk assessment is where most people fail with AI. You can't manage what you haven't inventoried.
Robert, that’s a fair point. We actually pushed a "shadow AI" policy update last week to ban the input of PII into public LLMs. Patricia, your advice on expanding the "Third-Party Risk" section is gold. Most of our AI risk comes from the vendors we're already using who just added "AI features" overnight. I'll focus on updating our vendor due diligence questionnaires to include model transparency and data retention clauses, keeping it all under the existing GRC umbrella.