Agentic AI and Model-To-Model Attacks COASP Practice Questions (Page 6)

Questions 26–30

1. 26expert · hard

   A security architect is designing an agentic AI system that uses a public LLM for planning and a private model for executing financial trades. The system must minimize the risk of the public LLM being manipulated to cause unauthorized trades. The architect has the following constraints: the public LLM is required for its advanced reasoning, and the private model must be able to execute trades quickly. Which mitigation strategy best balances security and performance?

   • AAllow the public LLM to directly call the private model's API
   • BUse a rule-based filter to validate trade instructions before execution
   • CAdd a human-in-the-loop approval for every trade
   • DReplace the public LLM with a smaller internal model
   Reveal Solution Discussion Select an answer first
2. 27application · medium

   A security team is conducting an attack surface analysis of an agentic AI system that includes a planning model, a memory module, and a web search tool. Which of the following is a potential attack vector that is unique to agentic systems?

   • AStealing the training data of the planning model
   • BUsing a denial-of-service attack to overwhelm the planning model
   • CExploiting a vulnerability in the web search tool's API
   • DManipulating the memory module to inject false information that influences future decisions
   Reveal Solution Discussion Select an answer first
4. 28foundation · easy

   What is the purpose of output filtering in mitigating model-to-model attacks?

   • ATo increase the diversity of the model's responses.
   • BTo reduce the computational cost of running the model.
   • CTo improve the model's accuracy by removing low-confidence predictions.
   • DTo detect and block malicious or unintended content generated by the model before it is used downstream.
   Reveal Solution Discussion Select an answer first
5. 29foundation · easy

   Which characteristic is a defining feature of agentic AI compared to traditional AI systems?

   • AIt autonomously plans, uses external tools, and maintains memory to achieve a goal.
   • BIt is limited to processing static datasets and cannot interact with the environment.
   • CIt requires human intervention for every decision and action it takes.
   • DIt uses a single large language model to generate responses without external tools.
   Reveal Solution Discussion Select an answer first
6. 30foundation · easy

   Which component is essential for an agentic AI system to remember past interactions and use them in future decisions?

   • AActivation function
   • BMemory
   • CTokenizer
   • DEmbedding layer
   Reveal Solution Discussion Select an answer first