Every company introducing AI runs into the same uncomfortable question: who, exactly, is the AI? When an assistant sends a mail, books a meeting or edits a file, something has to authenticate. Most tools answer with a service account — a synthetic user with broad standing permissions and no seat in the org chart.
That answer breaks two things at once. Access control breaks, because the bot's permissions become the union of everything anyone might ever need. And accountability breaks, because the audit log now says a robot did it — and the robot reports to no one.
worX answers differently: the agent is you.
Your permissions are the boundary
worX agents sign in silently through your existing Windows business account. Whatever Microsoft 365 lets you see is what the agent can see — no more, no less. A colleague's confidential folder stays exactly as closed to your agent as it is to you.
That single design decision removes the scariest failure mode of enterprise AI: an over-permissioned bot quietly reading everything.
Approval before every write
Reading is one thing; acting is another. Every sensitive action an agent wants to take — sending a reply, moving a file, changing a record — pauses and asks first. Approvals are durable: they survive restarts, they are logged, and every run is idempotent, so the same operation is never processed twice.
Accountability that auditors recognise
Because the agent acts as you, the trail it leaves is one your organisation already knows how to read: your identity in the audit log, your approval on the action, your permissions as the boundary. AI stops being a parallel system of record and becomes a very fast colleague who always asks.