Agent Risk Taxonomy and Minimal Footprint - Interactive Visualization

Agent risk is best understood through two dimensions: reversibility (can the action be undone?) and blast radius (how many users or systems are affected?). Actions in the irreversible + widespread quadrant - like deleting a production database or posting publicly - are critical risks requiring strict human oversight. The minimal footprint principle holds that agents should request only the permissions they need, prefer reversible actions, confirm before irreversible steps, and scope resource access tightly. Threat scenarios like prompt injection, data exfiltration, and unintended deletion each require specific architectural mitigations beyond general caution.

• Reversible + local actions (read file, write log) are low risk and can be fully automated
• Irreversible + widespread actions (delete database, public post) are critical risk and require human approval
• Minimal footprint means requesting fewest permissions needed and preferring reversible alternatives whenever possible
• Prompt injection, data exfiltration, and unintended deletion are the three highest-priority threat classes in agentic systems

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.