SSRF Through AI Tool Calls COASP Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   How does network segmentation help prevent SSRF in AI systems?

   • ABy encrypting all traffic between the AI and external services.
   • BBy using a firewall to block all inbound traffic to the AI system.
   • CBy blocking all outbound traffic from the AI system.
   • DBy isolating the AI system's network from sensitive internal services, so even if SSRF occurs, the attacker cannot reach them.
   Reveal Solution Discussion Select an answer first
2. 22expert · hard

   A large enterprise has an AI assistant that can fetch internal documents from a document management system. The assistant runs in a shared VNet with other applications, including a legacy database that contains sensitive financial data. The security team must mitigate SSRF but cannot change the legacy database's network location. Which of the following is the most effective mitigation?

   • ADisable the document fetching tool entirely.
   • BImplement a strict allowlist of document management system URLs and apply a NetworkPolicy that restricts the assistant's egress to only that system's IP.
   • CUse a web application firewall (WAF) to filter malicious URLs.
   • DMove the AI assistant to a separate VNet with no peering to the legacy database's VNet.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A company is deploying an AI assistant that can fetch web pages to answer user questions. They want to prevent SSRF. Which of the following is the most robust approach?

   • ADisable the fetch tool entirely.
   • BUse a dedicated egress proxy that enforces an allowlist of permitted domains and blocks private IP ranges.
   • CUse a URL shortener service to hide the final URL.
   • DAsk the AI model to only fetch URLs that look safe.
   Reveal Solution Discussion Select an answer first
5. 24application · medium

   An AI assistant is integrated with a tool that fetches user-supplied URLs to generate link previews. The development team wants to prevent SSRF. They decide to validate the URL by checking that the hostname resolves to a public IP address before making the request. Which of the following is a known weakness of this approach?

   • APublic IP addresses are often blocked by corporate firewalls.
   • BThe URL may contain a username and password that bypasses validation.
   • CThe AI model may refuse to fetch URLs that are not in its training data.
   • DDNS rebinding can make a public IP resolve to an internal IP during the actual request.
   Reveal Solution Discussion Select an answer first
6. 25application · medium

   A company's AI agent has a tool that fetches internal metrics from `http://metrics.internal/api`. The tool accepts a `metric` parameter and constructs the URL as `http://metrics.internal/api/{metric}`. A security review recommends mitigating SSRF. Which control is the most effective and least likely to break functionality?

   • AValidate the `metric` parameter against a list of allowed metric names.
   • BBlock the `metrics.internal` domain from being fetched.
   • CAdd a query parameter to the URL to prevent caching.
   • DUse a longer timeout for the HTTP request.
   Reveal Solution Discussion Select an answer first