Security

Why permission-scoped agents beat shared service accounts

Enterprise teams often connect an AI assistant through one shared bot identity: a service account with broad access to Jira, Slack, Drive, or a CRM. That pattern is convenient for demos and dangerous for real work.

The shared-bot failure mode

A shared service account collapses three different questions into one credential:

  • Who is asking?
  • What are they allowed to see?
  • Who is accountable for the write that followed?

When those answers disappear into a common bot, security reviews stall, audits get harder, and people stop trusting consequential actions.

What permission-scoped means

In a permission-scoped model, each teammate authorizes their own connected accounts inside the active workspace. Retrieval and write actions run through those identities rather than a shared all-access principal.

That design has practical consequences:

  • Private channels and docs stay private to people who already had access
  • Tool calls can be inspected against a real user and workspace boundary
  • Expanding a pilot does not require inventing a new superuser for every system

What to ask vendors

Before you approve an agent that can update tickets or post in Slack, ask:

  1. Whose identity performs the read?
  2. Whose identity performs the write?
  3. Can a user see the inputs, outputs, failures, and source links for a run?
  4. How is stored content encrypted and scoped to a tenant?

If the honest answer is “a shared bot does everything,” you do not have an operating model yet. You have a demo path.

Pipeer is built around tenant- and user-scoped access, isolated OAuth identities, and inspectable agent activity for exactly this reason.