Boundary Studio — autonomy boundary configured
Credit actions may proceed autonomously only up to £500,000 when the configured Watchtower conditions allow it.
PUBLIC MVP DEMO
Runtime control and evidence for agentic AI in regulated financial workflows.
This demo shows ACE intercepting a proposed AI-assisted credit action, evaluating it against configured autonomy boundaries, blocking execution where authority is unbound, and producing a verified per-decision evidence record.
01 — THE WORKING MVP
The recording walks through the local ACE MVP surfaces: Boundary Studio-lite, ACE Gateway, and ACE Evidence Console.
02 — WHAT THE REVIEWER SHOULD SEE
Credit actions may proceed autonomously only up to £500,000 when the configured Watchtower conditions allow it.
The test scenario asks ACE to evaluate a credit decision-support agent attempting a £650,000 action.
The action exceeds the autonomous boundary and no accountable actor is bound, so ACE blocks execution.
The reviewer retrieves the sealed per-decision record showing boundary, authority, execution, policy, and trace fields.
The Console recomputes the SHA-256 trace hash and confirms the sealed record still matches the canonical payload.
03 — CURRENT STATUS
ACE is currently a working local MVP. It is not yet a production deployment. The current demo uses local SHA-256 trace verification, SQLite storage, policy hash binding, and a simulated credit decision-support workflow. Planned extensions include RFC 3161 timestamping, institutional signing, role-directory validation, authentication, and production deployment hardening.