You ask an AI (artificial intelligence) assistant to help with real work. Maybe it drafts replies or books appointments. Then someone suggests the fastest setup: just let it sign in as a staff member.

That sounds easy. It also makes responsibility harder to track.

If an acting system uses a person’s login, you lose a clear answer to a basic question: who did this, the employee or the assistant? Once an assistant can take actions, the login it uses becomes the first safety check.

Give the assistant its own login before it can act

If an AI assistant can take actions for your office, it needs a clear login of its own, with its own permissions. Later guidance from the National Institute of Standards and Technology (NIST), a US government standards agency, supports that approach as the right foundation for acting systems.

That means an AI assistant should not sign in under an employee’s name. It should use a separate login created for the work it is allowed to do. Then you can see what it did, limit what it can reach, and turn it off without locking out a real person.

In our experience, the part people miss is how quickly a “temporary” shared setup becomes the normal setup. Once that happens, nobody can easily tell whether a change came from a person or the acting system. Undoing a bad change later, sometimes called a rollback, also gets harder.

NIST also points out another problem. An assistant carries the trust of the place that created and runs it. That includes how its keys are handled. A key is the secret or digital proof a system uses to show it is really itself. If the assistant works across more than one organisation or service, that chain of trust gets longer.

For an ordinary office, the plain English version is simple. If the assistant acts under Sam’s name, everyone downstream has to trust Sam’s account and the assistant behind it. That is too blurry for work that affects schedules or approvals.

A separate login also helps with consent. The World Wide Web Consortium (W3C), the standards group for the web, describes stronger sign-in methods that use public key credentials. Those are login credentials tied to a specific site and a device or security tool, rather than a reusable password. Their draft also says actions should not happen without user consent. That lines up with a practical office rule: sensitive actions need a human step, even when an assistant does the routine part first.

Think about a small dental practice. An assistant may be useful for drafting reminder messages or preparing a daily schedule. It should not quietly use the front desk worker’s login to change account settings or approve something in that person’s name.

Check these settings before you approve any AI workflow

  • Ask which login the assistant will use. If the answer is an employee account, stop there and ask for a separate login for the assistant.
  • Check what it is allowed to do. Give it only the permissions needed for the specific task, not broad access to everything that person can reach.
  • Make sure actions can be traced. You should be able to tell which changes came from the assistant and which came from a human.
  • Require a human approval step for sensitive actions. Drafting is different from sending. Preparing is different from approving.
  • Confirm you can turn it off fast. Removing the assistant should not disable an employee’s own access.

Common questions still lead back to separate access

Can an assistant use a shared office login if it is easier?

No. A shared or borrowed login hides who acted and makes review harder. A separate login for the assistant keeps responsibility clear.

Does every assistant need the same access as the person who requested it?

No. It should get only the access needed for the task. If it needs more later, add that deliberately instead of copying a person’s full access.

What if the assistant only works inside one department?

The same rule still helps. Even inside one office, a separate login makes it easier to see its actions and remove access without affecting staff.

Is a strong sign-in method enough on its own?

No. A stronger sign-in method helps prove who is signing in, but you still need clear permission limits and a way to remove access quickly.

Set the rule before anyone builds the workflow

Before you approve any AI workflow, decide one thing first: no assistant may act under a person’s name. Put that rule in writing before setup starts, so every request begins with a separate login and named approval points for sensitive actions.

Where does this advice come from?