Skip to content

Compliance officers

AI tools that send client data to unknown places, with no record of what they did, are a supervision problem. Keystile puts every AI action through your firm’s policies first.

For compliance: approvals and a record you can verify · 0:44
Question Where the answer is
What can AI tools do here, and who decided? The rules and tool policy on the Policies screen, and the record of who proposed and approved each change
Does a person approve anything sensitive? Approvals: held actions and who decided each
What did the AI actually do? The signed record, exportable as a spreadsheet; Verify the record proves it was not altered
Where does client data go? Your connected apps only. Client identifiers are masked or stopped before they reach an AI tool
Can staff use AI outside these controls with firm data? AI tools reach firm apps only through Keystile’s one central key; nobody gives an AI tool an app password
Where is Keystile hosted? On your firm’s own servers or in your own cloud account, if you choose
  • Rules in plain English, judged by meaning, not keywords (Rules).
  • Exact limits: payment caps, recipient allowlists, “a person approves every time” (Tool limits).
  • Four eyes on every policy change: one person proposes, a different person approves.
  • Tasks: an agent works on one client’s matter with only the actions that job needs (Tasks).
  • Protection level: Keystile Fortress applies both the fixed checks and Keystile Intelligence (Levels).

Use Replay on a few weeks of past agent logs, or run Shadow mode beside one workflow, to see exactly what would have been held or stopped — and the false alarms — before anything is enforced.

On the console’s Activity tab, filter by held or stopped and the last 7 days to see every decision with its rule and, for held actions, who approved or rejected it. Export it as a spreadsheet for your files. Filter by an app such as LinkedIn or YouTube to see every public post and who approved it.