Question 500
While investigating a FastMCP security review for fictional Elmcrest Theatre, the team reports: A retrieved configuration note includes a disguised URL that would leak network inventory if the agent automatically opens it. Which targeted correction best addresses the network automation failure?
Explanation
Correct Answer
Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints. This addresses the problem because tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. Confirm success using: A simulated malicious tool response cannot trigger outbound requests to unapproved destinations or reveal hidden credentials. Applied here, the relevant facts are: A retrieved configuration note includes a disguised URL that would leak network inventory if the agent automatically opens it. The question asks which targeted correction best addresses the network automation failure.
Incorrect Answers
The proposed action belongs to the mechanism: aI recommendations are hypotheses requiring external ground truth; absent evidence is not proof of a network condition. Here the stated fault follows a different mechanism: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The corrective choice is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.
This alternative is a response to another technical cause: agent tool chaining can enlarge the blast radius of an ambiguous user instruction unless execution permissions are independently gated. The reported case instead depends on: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The action that directly addresses it is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.
The proposed action belongs to the mechanism: aI-generated code may embed third-party patterns or dependencies; tool output is not automatically cleared of licensing obligations. Here the stated fault follows a different mechanism: tool results are untrusted external input; unrestricted outbound connectivity turns an injected instruction into a data-exfiltration path. The corrective choice is “Restrict tool egress to approved destinations and never grant untrusted retrieved text authority to choose exfiltration endpoints”.