Answers
What can an AI assistant see in my strategy?
The question has no answer until you know whose eyes the assistant borrows. In Empiraa GPS it borrows yours: the connection acts as the person asking, so it reaches what they already reach and nothing more. If you cannot open something in GPS, you cannot get it out through an assistant either, however the question is phrased.
- Set up with your plan
- Unlimited users
- 30-day pilot
Why the question is hard to answer
Three products can say "connected" and mean three different things.
The word covers arrangements that behave differently on an ordinary day and very differently on the day somebody leaves. Worth establishing which one you are being sold.
What "connected" can mean
The assistant gets its own account
A login or key belonging to the assistant, usually with wide access because narrowing it is work. Nobody owns it, it does not change when a person changes role, and it is the hardest arrangement to reason about a year later.
The assistant acts as the person asking
No access of its own. Every request is made with the asker’s permissions, so the answer to what it can see is simply what they can see. This is what GPS does.
A scoped token for a service
Narrower than a full account and still separate from any person. Reasonable for a scheduled job that runs with nobody present, and a poor fit for an assistant answering a question somebody just asked.
How it works in GPS
Your role is the boundary, and an administrator sets the ceiling.
How it works
It acts as you, so the answer is already known
Because the connection carries your permissions rather than its own, you do not need a new mental model for what an assistant can reach. It is the plan you already open. A person who cannot see a particular objective in GPS cannot have it read out to them by an assistant, and an administrator who connects one is connecting administrator-level sight, which is worth knowing before they do it.- No separate account, key or service identity for the assistant
- Your existing role decides what it reaches
- Access changes the moment your role does
An administrator sets Off, View only or Full
One setting for the whole organisation rather than a per-person negotiation. Off means no assistant reaches the plan at all. View only means anyone may ask what is off track, who owns a goal or what changed this month, and nothing can be altered. Full allows changes, within each person’s own permissions. Most organisations start on View only and move once they have watched it for a cycle.- Organisation-wide, not per seat
- View only is a complete answer to "can it break something"
- Changeable at any time, in both directions
Anything it changes is marked, previewed and reversible
Reading is only half the worry. Every change made through an assistant is labelled as made through an assistant, so a month later you can tell whether a person or an assistant recorded a number. Bulk changes are previewed before they save rather than applied and reported. Most changes can be undone for seven days. And protective limits can pause the connection, leaving it for the owner to resume.The part that is not ours to promise
Whatever the assistant reads lands in that conversation, which sits with the assistant vendor rather than with us. So there are two data questions, not one: what GPS lets the assistant reach, which this page answers, and how Claude or ChatGPT handles the transcript afterwards, which is set by your agreement with them. Anyone selling you an MCP without mentioning the second question has left out the half you cannot configure from our side.What to settle before you switch it on
Six questions, answerable in an afternoon.
None of them are technical. They are the ones a cautious executive asks, and having answers ready is usually what unblocks the decision.
Before you connect
Whose permissions does it use?
If the answer is anything other than the person asking, find out who holds the access instead, and what happens to it when that person leaves.
Who in the organisation will connect it?
Because it acts as them. An administrator connecting an assistant is a different exposure from a team lead doing it, with no setting changed.
Which setting do you start on?
View only for a cycle is the cheap answer. It gives you the benefit people actually notice, which is being able to ask, with nothing at risk.
Is a change traceable to the assistant?
It should be marked on the record rather than inferred from a timestamp. Without that you cannot audit the quarter afterwards.
How long is the undo window?
Days is reasonable. In GPS most changes can be undone for seven. No window means relying on nobody making a mistake.
What does your assistant vendor retain?
A separate contract from ours, covering the conversation rather than the plan. Read it once, at the start, rather than during an incident.
Questions
Questions people ask about assistant access
What can an AI assistant see in my strategy?
In Empiraa GPS, exactly what the person asking can see. The connection acts as that person rather than holding access of its own, so their existing role is the boundary. An administrator can additionally set assistants to Off, View only or Full for the whole organisation, which sets the ceiling for everybody.
Does the assistant get its own login to our data?
No. There is no separate account, key or service identity for the assistant in GPS. That matters most on the day someone changes role or leaves, because access that belongs to a person changes with them, while access that belongs to an assistant tends to outlive everyone who remembers granting it.
Can an assistant see things I am not allowed to see?
No. Because every request is made with your permissions, anything you cannot open in GPS is equally out of reach through an assistant, and no phrasing of the question changes that. The flip side is worth noting: whoever connects it is lending their own level of sight.
Can we let it read the plan but never change it?
Yes. View only is an organisation-wide setting an administrator chooses, and it is where most organisations start. People can ask what is off track, who owns a goal or what changed this month, and nothing in the plan can be altered through an assistant while it is on.
How do we know whether a person or an assistant changed something?
Every change made through an assistant is labelled as made through an assistant, so the record itself tells you rather than leaving you to infer it. Bulk changes are previewed before they save, and most changes can be undone for seven days.
Where does our strategy data actually go?
What the assistant reads is used to answer the question in that conversation, which sits with the assistant vendor rather than with Empiraa. So there are two questions to settle: what GPS allows the assistant to reach, which is set by the role and the organisation setting, and how your assistant vendor handles the conversation, which is governed by your agreement with them.
Can we turn it off again?
Yes, at any time, for the whole organisation, and it takes effect immediately rather than at the end of a billing period. Protective limits can also pause the connection on their own, leaving it for the owner to resume once they have looked at why.
Still working out the access model? Ask ANI and get a straight answer.
Start on View only. Decide later.
Start free, set the organisation setting before anyone connects anything, and move it when you have watched a cycle.