Agent Hijacking COASP Practice Questions (Page 5)

Questions 21–25

1. 21application · medium

   A financial AI agent is authorized to transfer funds between internal accounts. To reduce the risk of hijacking via prompt injection, which access control measure would be most effective?

   • AAllow the agent to use any internal API without restrictions
   • BRequire a human approval step for any transfer above a small threshold
   • CDisable logging of all agent actions to reduce complexity
   • DStore all API credentials in the agent's system prompt
   Reveal Solution Discussion Select an answer first
2. 22foundation · easy

   What is a potential consequence of agent hijacking on data integrity?

   • AThe agent's model weights may be copied by the attacker
   • BThe agent may modify or corrupt data it accesses, leading to inaccurate records
   • CThe agent's user interface may become unresponsive
   • DThe agent's response time may increase due to extra processing
   Reveal Solution Discussion Select an answer first
4. 23foundation · easy

   Which of the following is a likely impact of agent hijacking on user trust?

   • AUsers may see an increase in the agent's accuracy
   • BUsers may receive more personalized recommendations
   • CUsers may lose confidence in the agent's outputs and stop relying on it
   • DUsers may experience faster response times due to optimized processing
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   An AI agent that manages a company's inventory is suspected of being hijacked. Which combination of monitoring activities would provide the most comprehensive detection?

   • ACheck the agent's model version weekly
   • BLog all agent actions and compare them against a baseline of normal behavior
   • CMonitor only the agent's response time
   • DReview the agent's training data for changes
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A security analyst is explaining agent hijacking to a colleague. Which statement best defines agent hijacking?

   • AAn attacker gains physical access to the server hosting the agent
   • BAn attacker manipulates an AI agent's inputs or environment to control its behavior or outputs
   • CAn attacker replaces the agent's underlying model with a malicious one
   • DAn attacker steals the agent's training data
   Reveal Solution Discussion Select an answer first