Exams / AIF-C01 / Securing AI systems / Question 4
AIF-C01 practice question: Securing AI systems (4 of 5)
Security, Compliance, and Governance for AI Solutions A free question from our original AIF-C01 practice exam. Try it, then open the answer to see why each option is right or wrong.
An e-commerce company's AI agents call tools through Amazon Bedrock AgentCore Gateway. Security wants a deterministic rule that refund tools can only be called for amounts under a set limit, enforced outside the agent's own code so a manipulated agent cannot bypass it. Which capability fits?
Choose one answer.
- AA. Policy in AgentCore, evaluating tool calls at the Gateway
- BB. A sentence in the agent's system prompt telling it to respect the limit
- CC. A Bedrock Guardrails word filter that blocks the word "refund"
- DD. AWS Trusted Advisor checks run against the agent's account
Show the answer and explanation
Answer: A
- A. Right: Correct. Policy intercepts agent traffic at the Gateway and evaluates each request against deterministic rules, outside the agent's code.
- B. Wrong: Instructions in a prompt are not deterministic, and a manipulated agent could ignore them.
- C. Wrong: A word filter blocks text in prompts and responses. It cannot enforce a numeric limit on tool parameters.
- D. Wrong: Trusted Advisor reviews an account against best practices. It does not evaluate individual tool calls.
Explanation
Policy in Amazon Bedrock AgentCore puts a security boundary around tool calls. Policies are written in Cedar, or authored in natural language and then validated, and are associated with a Gateway. Every tool request is checked against them, scoped by user identity and tool input values, and each decision is logged. AgentCore Identity handles who the agent is and what credentials it uses; Policy controls what it is allowed to do.
Study
Keep practising
Answer questions with a score at the end, or practise just this topic with the explanation after each one. No sign-up needed.