Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Agent permissions

Agent permissions are the rules that decide which tools, files, commands, networks and accounts an AI agent may use, and which actions need a human's approval first, so a mistaken or manipulated agent can do limited damage.

Also called: least privilege, allowlist, scoped access, principle of least privilege

By Divya Sharma · Checked Nov 15, 2026 · How we review

An agent's risk is set less by how clever the model is than by what it is allowed to touch. A model that only writes text can say something wrong. An agent with shell access, a production database login and a company email account can act on that mistake. OWASP calls this risk excessive agency: damaging actions taken in response to unexpected, ambiguous or manipulated model output, whatever the cause.

Least privilege

The principle of least privilege is an old security rule. NIST defines it as restricting the access of users, or processes acting for them, to the minimum needed for their tasks. For agents, OWASP applies it in layers: offer only the tools the task needs, give each tool only the functions it needs (a summariser needs to read email, not send it), and give each tool's credentials only the scopes it needs.

How it looks in practice

Coding agents let you allowlist commands that run without asking, deny others outright, and require approval for the rest. File access can be limited to the project folder. Network access can be blocked or restricted to listed domains, ideally inside an agent sandbox. Account access is best granted through scoped tokens, for example via OAuth for agents, instead of a broad API key. Sensitive steps such as payments, deletions or messages to customers go through human-in-the-loop approval.

Why it matters

Permissions are the backstop when other defences fail. If a prompt injection gets through, narrow permissions decide whether it can do real harm. Before installing an MCP server, check what it asks for; each MCP tool review on this site lists the permissions requested.

Related terms

Human-in-the-loop
Human-in-the-loop (HITL) is a design in which an AI system pauses at chosen points for a person to approve, correct or reject its next step, so risky or uncertain actions are not taken automatically.
Agent sandbox
An agent sandbox is an isolated environment, such as a container, virtual machine or operating-system restriction, in which an AI agent runs code and commands so it cannot reach files, networks or data outside set limits.
OAuth for AI agents
OAuth for AI agents is the use of the OAuth authorization standard to let an agent or MCP client act on a user's behalf with a limited, revocable access token, instead of holding the user's password or a broad API key.
API key
An API key is a secret string an application sends with each request so that a service can tell which account is calling, apply its limits and bill it. Anyone holding the key can usually use it.
Prompt injection
Prompt injection is an attack where text fed to a language model, typed by a user or hidden in a web page, file or tool result, changes the model's behaviour in ways its developer did not intend.

Sources