Agentic risks
Also known as: agent risk · autonomy risk
Agentic risks are the safety problems that appear when an LLM acts — calling tools, editing files, making decisions — rather than just answering.
- A wrong tool call with real side effects.
- An exfiltration path through a tool.
- A loop that burns cost or hammers an API.
- Decisions that are hard to audit after the fact.
The shift: a chatbot's worst case is a bad answer. An agent's worst case is a bad *action*. The risk scales with the agent's privileges — so keep them minimal.