AI UX PlaygroundNewsletterJoin 2K+ AI designers and PMs on Substack. New teardowns, patterns, and prompts as they drop.

Trust

Granular Consent

Split permissions into per-capability toggles (read vs send mail, view vs book calendar) instead of one all-or-nothing OAuth blob. People revoke what they do not need without tearing down the whole integration.

Interactive demo

Inbox assistant

Allowed actions

Turn each action on or off. Changes apply immediately.

2 actions turned on

Read inbox

Summarize threads and detect action items.

Send email

Draft and send as you, including external domains.

Modify calendar

Create and move events on your primary calendar.

Overview

How might we design granular consent so people can trust and act on AI output?

When to use

  • Essential for AI assistants, enterprise agents, and connector ecosystems where least-privilege permissions reduce risk and let users tailor scope to the specific task without losing the rest of the integration.

When to skip

  • Tiny prototypes with a single harmless capability where toggles add clutter.
  • Environments that only support coarse OAuth scopes and cannot honor fine toggles.
  • One-shot local tools with no persisted permissions.

Rules

  • A single “Allow all” default that collapses granularity.

  • Toggles that look independent but still require re-auth of everything on change.

  • No plain-language description of what each capability can do.

  • Hiding previously granted scopes so users cannot audit them later.

Evidence

ProductImplementation
GitHub AppsFine-grained repository and permission scopes at install time.
Google Workspace add-onsPer-API scopes rather than blanket account access.
iOS / macOS privacy panesPer-capability access to photos, mic, and screen recording.
ChatGPT / Claude connectorsConnector permissions broken into discrete actions where platforms allow.

Real-world examples

See all

FAQ

What is granular consent for AI agents?

Granular consent lets users grant or revoke individual agent capabilities instead of one bundled allow-all permission, matching least-privilege security to how people think about risk.

How does granular consent relate to scope disclosure?

Scope disclosure is the plain-language explanation of what the agent can reach; it is part of Granular Consent, not a separate pattern. Pair the explanation with per-capability toggles, default expiry, one-click revoke, and a drift signal when permissions accumulate.

Why avoid bundled agent permissions?

Bundled grants over-authorize “just in case,” expand blast radius on mistakes or compromise, and make revoke all-or-nothing. Per-capability toggles keep access tied to the current task.