SSRF Through AI Tool Calls COASP Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   Which of the following is an effective mitigation against SSRF in AI tool calls?

   • AUsing a blacklist of known malicious URLs.
   • BEncrypting all traffic between the AI and external services.
   • CImplementing an allowlist of permitted domains or IP addresses for tool calls.
   • DDisabling all network access for the AI system.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   An AI agent is designed to fetch internal documentation from a corporate wiki. The agent receives a user request and constructs a URL by concatenating a base path with a user-provided filename. An attacker submits a filename like '../../../../etc/passwd'. What is the most effective mitigation to prevent this attack?

   • AValidate the final URL against an allowlist of permitted domains and paths before making the request.
   • BUse a longer base path so that the traversal cannot reach the root directory.
   • CEncode the filename using URL encoding before concatenation.
   • DSanitize the filename by removing any occurrence of '..' before constructing the URL.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   An AI assistant has a tool that fetches a webpage and summarizes it. The tool takes a `url` parameter. A user submits the prompt: "Summarize http://192.168.1.1/secret". The tool fetches the URL and returns the summary. Which component is the primary enabler of this attack?

   • AThe user's ability to craft a prompt that includes a URL.
   • BThe server's firewall, which allows outbound connections to internal IPs.
   • CThe AI model's natural language understanding, which interprets the prompt as a request to fetch the URL.
   • DThe tool's lack of validation on the `url` parameter, allowing it to fetch any URL.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   In an AI agent architecture, how does the agent typically invoke an external tool?

   • AThe agent sends a structured request, such as a JSON object with a tool name and parameters, to a backend function that executes the tool.
   • BThe agent sends an HTTP request directly from the user's browser to the external tool's API.
   • CThe agent uses a plugin that runs entirely on the client side.
   • DThe agent directly executes shell commands on the server without any intermediary.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A company is deploying an AI agent that can fetch URLs to retrieve real-time stock prices. The agent runs in a VNet with access to internal monitoring systems. Which combination of controls would most effectively prevent SSRF while preserving the agent's functionality?

   • AAllow all URLs but log all requests for later review.
   • BUse a deny list of internal IP ranges and rely on the AI model to refuse malicious requests.
   • CImplement a URL allowlist for external financial APIs and place the agent in a separate subnet with no route to internal systems.
   • DRestrict the agent to use HTTPS only and disable redirects.
   Reveal Solution Discussion Select an answer first