Policy Bootstrap.
When we’ve seen enough.

WhiteFin sits inline on every agent-to-tool call. Nothing is blocked. Every call is recorded, labeled, and grouped. Your agents continue to run exactly as they did before; the only difference is that for the first time, you have ground truth about what they actually do.

Observation does not end on a date. It ends when an agent stops showing us anything new — enough evidence behind every distinct behaviour, then a sustained run across multiple sessions in which no new behaviour appears. A busy agent gets there in days; a periodic one takes longer. Either way the answer comes from what we have seen, not from a calendar.

A proposed policy set is generated from everything observed. We sit with your team and walk through every rule. You see — in plain English, against real recorded traffic — what each rule allows, what it would have blocked, and what it would have escalated. You approve, edit, or reject before anything ships.

You flip the switch. Deny-by-default goes live. Approved tool calls pass. Unapproved calls stop, or route to a human approval queue, depending on how you set the rule. There are no surprises, because the rules are the ones you approved.

Some agents keep broadening for as long as you watch them. Rather than wait forever, WhiteFin caps observation at a ceiling you configure and enforces against what was learned — and it labels that agent CONTAINED, not understood. Nothing it has never done is allowed: deny-by-default covers the rest. You get told which of your agents we know and which we are merely holding, because those are different claims and only one of them should let you sleep.