Acutis Gate logo Acutis Gate identity for AI at work

How to stop an AI agent from using a shared admin account

Short answer: connect the AI to your apps through a gateway that carries each person's own key or sign-in to the app, instead of one service account everyone shares. Then the app's own permissions decide, exactly as they do when that person logs in themselves. For the few older systems that only take one account, keep it, but label every call made with it and fence it with rules.

How the shared account sneaks in

The quickest way to connect an AI to the HR system or the CRM is one API key with wide access, pasted into one tool server. Everyone's AI then reaches everything that key can, whatever their own role is. The HR system's logs show one account doing everything, so "who looked at the salary file?" has no answer.

A quick check: ask the AI the same question as an HR partner and as someone in Finance. If they get the same answer, your AI is running on a shared account.

Three ways to give the AI the person's own access

  • Each person's own key: every employee stores their own API key or token for the app, once. Their AI uses it, so the app sees them.
  • Each person's own sign-in: for apps that accept the same sign-in token, the person's token is passed straight through.
  • Windows sign-in: for file shares on the on-prem editions, the call runs with the person's own Windows identity (Kerberos), under the NTFS and AD rights you already set.

Set it up in Acutis Gate

  1. As an admin, open My apps → Add an app for everyone, enter the app's MCP address, and choose how people sign in to it: Each person's own key, Each person's own sign-in, One shared account (logged as shared) or No sign-in needed.
  2. For "own key" apps, each person opens My apps and adds their own key. Nobody else can see or use it, admins included. Keys are encrypted at rest and can carry an expiry date.
  3. Ask the AI something from that app. Until a person adds their key, Gate tells them so instead of falling back to someone else's access.

Every audit row records the identity the call actually ran as, so the difference is visible at a glance:

10:25:02Z  priya.n   hris__get_leave_balance   decision=allow  mode=delegated  ran_as=priya.n
10:31:47Z  priya.n   legacy__lookup_asset       decision=allow  mode=mapped     ran_as=service:legacy

When a system really only has one account

Some older systems have no per-person login. Gate can still connect them with One shared account: every call is marked "shared" in the trail, and rules decide who may use it at all, which actions are allowed, and which wait for an approver. That keeps the exception honest and small while you move the system to personal logins.

Every AI action, as the person who asked

Gate carries each employee's own access to HR, finance, the CRM and your file shares, and labels the exceptions.

Start a 14-day trial Tour the live Gate

Frequently asked questions

Why is a shared admin account risky for AI agents?

Everyone's AI gets the account's full reach regardless of their own role, and the app's logs show one account for everything, so you cannot tell who saw or changed what.

How does an AI agent act with the user's own permissions?

A gateway between the AI and your apps carries each person's own API key, sign-in token or Windows identity to the app, so the app's existing permissions apply to every call.

Can admins use an employee's stored key in Gate?

No. A personal key is stored for that person only and is used only for their own calls.

What if an app only supports one account?

Connect it as a shared account. Gate marks every such call as shared in the audit trail and lets you limit who may use it and which actions need approval.