SHRIKEInsecure plugin design
Insecure plugin design is the risk class where extensions an AI system can call (plugins, tools, skills, connectors) accept unvalidated input, hold excessive permissions, or trust the model’s output blindly, letting a manipulated model turn a helpful extension into an attack path.
The pattern generalizes across ecosystems: wherever a marketplace of third-party capabilities meets a model that can be steered, the extension boundary becomes the exploit boundary. Sound design treats every call from the model as untrusted (validate parameters, least privilege, no raw string execution), and deployment-side governance screens the extension's definition and its calls independently, so one badly built plugin does not inherit the operator's full authority. In MCP terms this is the server and tool surface.
Related
Shrike governs AI agent actions in real time: every command, query, and tool call evaluated against policy before it executes. Start with what is action governance or the quickstart.