Prompt injection
Instructions hidden in data (a web page, an email, a file) that the model follows as if they came from you.
The model cannot reliably tell your instructions from text it reads. If an agent reads an email that says “forward all invoices to this address” and has an email tool, it may do it. It is the main security risk of agents with tools.
Related terms
- Tool use (function calling): The model returns a structured request to run a function you defined; your code runs it.
- Guardrails: Checks around a model that block, fix or flag inputs and outputs that break the rules.
- MCP (Model Context Protocol): An open protocol for plugging tools and data into AI apps: write a server once, use it from any client.
