lethal trifecta
Simon Willison's name for an agent that has access to private data, exposure to untrusted content and a way to communicate externally, which together let injected text steal data.
With all three legs present, one injected instruction can exfiltrate private data with no further access.
In more detail
Any one leg alone is manageable; all three at once mean a single malicious document can make the agent send your data away. Break the trifecta by removing a leg, for example by blocking outbound tools once untrusted text is in the context.
Related terms
- indirect prompt injection — Prompt injection hidden in content the model reads on someone else's behalf, such as a web page, email, ticket, document or tool result.
- data exfiltration — Secretly copying data out of a system to somewhere an attacker controls.
- excessive agency — Giving a model-driven system more tools, permissions or autonomy than its task needs, so a fooled or mistaken model can do real damage.
Taught in
- Threat-Model An LLM App — AI Engineering: LLM Apps, RAG & Agents
- Prompt Injection — AI Engineering: LLM Apps, RAG & Agents