pydantic-ai

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The documentation shows the agent fetching and ingesting arbitrary public web content (e.g., references/agents.md "DocumentUrl/VideoUrl" and force_download, references/tools.md and references/models.md built-in WebSearchTool/WebFetchTool and Google YouTube/File upload examples, plus common tools like Exa/duckduckgo that retrieve web pages), which the agent is expected to read/interpret as part of its runs and could therefore allow untrusted third-party content to influence tool use and subsequent actions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The documentation includes configuring a remote MCP server URL (e.g., https://mcp.deepwiki.com/mcp) which is contacted at runtime to provide resources and tool responses that are injected into the agent's prompts/execution, so that remote URL can directly control prompts or execute actions during runs.