Enforcer
Your agents answer with everything they can reach, whoever is asking.
Enforcer is how we fix that: every answer limited to what the person asking is allowed to see. It is open to early design partners, so get in touch.
An agent shares its own access with whoever asks.
When an agent connects to SharePoint, the CRM or the warehouse with an account of its own, that account has to be broad enough to be useful. Every question anyone asks the agent is answered from it. The intern and the CFO are searching the same folders.
Each system is still enforcing its permissions correctly. It is enforcing them for the agent.
Every workaround costs you something.
Narrow the agent’s account
It leaks less. It also answers less, and people stop asking it things.
Keep it out of sensitive systems
The agent stays out of HR and finance, which are the systems people most wanted it for.
Sign in as each user
Some systems can pass the person through. An agent that spans several of them meets several definitions of who that person is.
Two people ask the agent about the same customer. They should get different answers.
The head of finance should see the contract value, the payment history and the open dispute. Someone in support should see the tickets and the contact details they already work with.
That is what Enforcer is for: each person gets the answer they could have found by looking themselves.
Put the agent where people want it.
When every answer fits the person asking, the agent can go into finance, HR and the CRM, and the security review has something it can say yes to.
The rollout stops waiting on the systems everyone was nervous about.
Scan, Watchdog, then Enforcer.
The Scan maps what your agents could reach before they go live, Watchdog checks what they do once they are live, and Enforcer controls what comes back.