CaMeL
A design from Google DeepMind that defeats prompt injection by having a privileged model write a plan as code while a quarantined model reads untrusted data, with data flow tracked and checked by policy.
In more detail
Debenedetti and colleagues published it in 2025. Its lesson for ordinary apps: put the decisive control in code that tracks where untrusted data flows, rather than in the prompt.
Related terms
- prompt injection — An attack in which text a model reads overrides or redirects the instructions it was given, because instructions and data share one channel.
- spotlighting — Marking untrusted text in a prompt, by delimiting, interleaving a marker or encoding it, so the model can tell it is data and not instructions.
- agent policy engine — Code outside the model that decides, for every proposed tool call, whether to allow it, deny it or ask a person, based on rules written as data.
Taught in
- Build An MCP Server — AI Engineering: LLM Apps, RAG & Agents
- Connect An Agent To MCP — AI Engineering: LLM Apps, RAG & Agents
- Prompt Injection — AI Engineering: LLM Apps, RAG & Agents