How We Built Security Into Custom Agents
- Source
- notion.com
- Author
- Notion Blog
- Date
Context
Notion designed Custom Agents, AI assistants that carry out tasks inside a team’s shared workspace, so their permitted actions are constrained by access controls outside the . Agents begin without access to most resources, and owners grant permissions such as viewing or editing particular pages. This limits what an can do even when an instruction or an external document steers it badly. The company’s security account describes a concrete lesson from testing: broad Slack write access let agents post in channels their owners had not intended. Notion introduced a narrower permission allowing replies only in threads that explicitly triggered the agent. It treats prompt-injection detection as an additional defense, while acknowledging that detection remains imperfect.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- tool permissions — The rules governing which tools an agent may call and which need confirmation — the boundary between a mistake and an incident.
- MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
Checking sign-in…
Loading comments…