The lethal trifecta for AI agents
by Simon Willison
A defensive framework for recognizing risky combinations of agent capabilities and untrusted inputs.
This security discussion identifies a risky combination of sensitive access, untrusted inputs and communication capabilities. We link the analysis for defensive design; no attack instructions or payloads are reproduced here.
- Level
- Advanced
- Models
- any
- Category
- Agents and harnesses
- Tags
- security, prompt-injection
- Source date
- 2025-06-16
- Added
- 2026-10-18
- Last verified
- 2026-10-18
- Status
- ok
Summary in our own words. All rights with the original creator. Correction or removal: DM on X.