liveBook · Manning

Plugin

As explained in AI Governance

A plugin is an add-on software component that extends an application or model-enabled system with additional capabilities or integrations.

Plugins and tools extend GenAI systems, but installation and granted permissions make them trusted parts of the software supply chain. A vulnerable or malicious component can become an entry point to data or networks without compromising the model itself. section 4.3.3

Common plugin-related attack paths include:

  • Flaws in installation or authentication can enable unauthorized plugin access, including zero-click access to private GitHub repositories. section 4.3.3
  • A malicious browser extension can steal session cookies and hijack business accounts. section 4.3.3
  • Trusted MCP servers can leak data across tenants or silently copy outgoing email to an attacker. section 4.3.3

A trusted adapter turns malicious

The package postmark-mcp impersonated a legitimate email service and operated as advertised for fifteen versions. A later version silently BCC’d outgoing email to an attacker-controlled address. section 4.3.3

How AI Governance covers it

How deep it goes
Traces installation trust into concrete compromise cases; broader MCP security for agentic systems is treated separately.
Examples
Uses AskTheCode repository access, malicious browser extensions, and compromised MCP services as attack examples.
What you can do after
Evaluate plugin trust by checking installation flows, granted permissions, and risks from inherited access.
How it connects
Connects plugins with tools, software supply-chain risk, authentication, and the Model Context Protocol.

Also explained in

Compare how 9 books explain plugin →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.