Which of the following mitigation control strategies would BEST reduce the risk of introducing hidden backdoors during model fine-tuning via third party components?
Answer(s): B
Performing threat modeling and integrity checks helps identify potential attack paths and verify the authenticity and integrity of third-party components, reducing the risk of hidden backdoors being introduced during model fine-tuning.
An organization is deploying a large language model (LLM) and is concerned that input manipulations may compromise its security. Which of the following is the MOST effective way to determine an acceptable risk threshold?
Answer(s): C
Assessing the business impact of known threats allows the organization to determine how much risk is acceptable based on potential operational, financial, regulatory, and reputational consequences, which is the correct basis for defining an AI risk threshold.
An organization is implementing AI agent development across multiple engineering teams. Which of the following is the MOST important focus of AI-specific security training for developers?
Answer(s): A
Prompt injection, agent memory control, and insecure tool execution are the most critical AI-specific security risks for developers building AI agents, as these directly affect how agents interpret instructions, retain context, and invoke actions, making them primary attack vectors that must be addressed at the development stage.
An organization is designing an AI-based credit risk assessment system that will integrate with sensitive financial data sets. Which of the following would BEST support the implementation of security by design principles in the AI system's architecture?
Applying threat modeling specific to AI components before deployment embeds security considerations into the system’s architecture from the outset, ensuring risks related to data, models, and integrations are identified and addressed early in line with security by design principles.
Which of the following types of data is used to tune hyperparameters?
Validation data is used to tune hyperparameters by evaluating model performance during training iterations, allowing adjustments that improve generalization before final testing.
Post your Comments and Discuss ISACA AAISM exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the AAISM content, but please register or login to continue.