AAISM: Advanced in AI Security Management

By Isaca · Question Mode

QuestionQ32

AI Risk Management

When strong input controls are not feasible for a large language model (LLM) to prevent prompt-injection attacks from external threats, which of the following is the BEST compensating control to mitigate the risk?

Explanation

Identity and access management applies authorization and least-privilege restrictions to the LLM's access to data, tools, and backend services. This limits the harm a successful prompt injection can cause, even when untrusted external input cannot be robustly controlled.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!