How IFS permissions govern an AI assistant
Why the IFS permission set is the right place for limits.
The short answer
The assistant can do what you can do in IFS, and no more.
How does IFS control access?
Through permission sets that give each projection a level: none, read-only, custom or full. Projections are the API, so the same grant governs every call an assistant makes.
What does that mean in practice?
The answer to what the assistant may do is not a policy document. It is the permission sets your IFS administrators already maintain, described in the IFS Cloud documentation.
What about data a person may not see?
It follows too. If a person may not see supplier prices, IFS does not return them, so there is nothing for the assistant to leak.
Your own login, or a shared account?
The design choice that decides most of a security review.
| Compared on | Signed in as the person | Shared service account |
|---|---|---|
| Who IFS sees | The person who asked | One integration user |
| Who enforces limits | IFS, through permission sets | The assistant’s own code and prompts |
| What IFS history shows | The person’s name on each change | Every change under one account |
What data goes to the model?
An assistant sends the request and the records it read to a language model. Ask where that model runs, what is kept, and where the rest is hosted.
Where does NgageChat run?
In your own cloud, so IFS traffic stays inside your environment, or run for you by EX10.
Which model, and what is kept?
Enterprise deployments run on Google Cloud Vertex AI with zero data retention configured, in the region you choose, and NgageChat is built to work with more than one model. If you have data residency duties, raise them in the first conversation: they shape the deployment.
More questions from IFS teams
Do people need a separate NgageChat login?
They sign in with their IFS identity. That identity is what makes the permission model work.
How does NgageChat sign people in to IFS?
With OpenID Connect against your IFS identity provider, using the authorisation code flow with PKCE. Each request then runs under that person’s own IFS permissions, never a shared service account.
What changes for our IFS administrators?
Less than they expect. Access is still decided by the permission sets they maintain today. The one new decision is who may use the assistant at all, an ordinary application access question.
Can an administrator see what people asked?
Conversation history is stored, and authorised administrators can review it. Treat that as a governance feature, and tell your users it exists.
What happens when someone leaves?
Disable their IFS account. Without a valid IFS identity there is no access token, and without a token there is no access to your data.
See NgageChat answer a question from your IFS
Send us a task you have not prepared for. We run it with you, live, on a test system.