AGENTIC AI CONTROL ENVIRONMENT · REGULATED FINANCE
ACE blocked execution because the action exceeded the boundary and no accountable officer was bound.
No accountable officer is bound. ACE blocks execution — and seals the proof. Veridom ACE is the runtime control and evidence layer for agentic AI in regulated financial workflows.
01 — THE CONTROL LOOP
A dashboard shows activity. A monitor shows performance. ACE shows whether the boundary held — for one specific action, at the moment it was attempted.
Approve a £650,000 facility.
Autonomous only to £500,000.
No accountable officer bound.
The action does not run.
Proof exists before anyone asks.
The action was never allowed to execute — and the institution can prove exactly why. That is the product: not the block, the proof of the block.
02 — SEE ACE INTERCEPT AN ACTION
The result is computed against a configured credit policy — autonomous to £500,000, assisted to £2,000,000 — exactly as the ACE Gateway evaluates it at runtime.
The action exceeds the autonomous boundary and no accountable authority is bound. ACE blocks execution — and still seals the record as evidence.
03 — THE PROOF RECORD
Search by decision ID, retrieve the sealed record, verify its integrity, export it. No access to any internal system required.
ACE-2026-0514-a3f1c9
04 — WHAT ACE DOES
CONFIGURATION
Risk, compliance, model-risk, and audit teams define autonomous limits, assisted thresholds, escalation conditions, hard-block rules, accountable-authority roles, and evidence requirements — turning policy intent into a machine-readable policy file the Gateway enforces.
RUNTIME ENFORCEMENT
Intercepts proposed agent actions before execution, evaluates them against the active policy, and returns a control result. If the action exceeds the boundary and no valid accountable actor is bound, ACE blocks execution — and still seals the record.
AUDIT & QUERY
Search by decision ID, retrieve the sealed record, verify integrity, export as JSON or Markdown. It answers one question: can this institution prove what happened for one specific AI-assisted action?
05 — THE ENFORCEMENT GATES
Each is a runtime gate with a trigger, an action, and sealed evidence. They may only move toward more control — never less.
Routes actions that exceed approved autonomy to more control.
Blocks execution when assisted or escalated actions lack valid authority.
Stops actions where sanctions, fraud, AML, or KYC flags are unresolved.
Prevents agents acting on missing, stale, or incomplete evidence.
Produces the decision-boundary proof record for audit, model-risk, or board review.
06 — BUILT ON OMP™
Published as Internet-Drafts at the IETF, with a permanent DOI on Zenodo. OMP separates three axes that must not be conflated.
An action can be routed to ASSISTED, fail authority binding, and be blocked — which is different from calling it "high risk" or "denied." ACE makes this state explicit, enforceable, and evidential. Read how OMP works →
07 — WHAT ACE IS NOT
ACE is a runtime control and evidence layer — it helps those functions prove whether a specific AI-assisted action stayed inside its approved boundary when it happened.
08 — CURRENT STATUS
The current build includes ACE Gateway, Evidence Console, Boundary Studio-lite, Watchtower configuration, policy export with hash binding, per-decision trace generation, integrity verification, and JSON/Markdown evidence export.
Production hardening — RFC 3161 timestamping, institutional signing, role-directory authority validation, policy approval workflow, deployment hardening, and additional financial-services Watchtower profiles — is on the near-term roadmap, co-developed with design partners.
09 — INTENDED USERS
Primary users: Head of Model Risk, Head of Audit Model Risk, CRO, Head of AI Governance, Internal Audit Lead, Compliance Officer, Financial Crime Control Lead.
ACE leads with one workflow: regulated credit decision-support. The same control loop extends to fraud review, payment-risk actions, KYC/CDD checks, financial-crime escalation, claims decision support, and internal control automation — but credit decision-support is the proving ground.
Veridom is a member of the NVIDIA Inception program. NVIDIA technologies are being evaluated for future ACE use cases — including agentic-workflow simulation, model-evaluation environments, and high-volume inference integration. ACE is model- and framework-agnostic: it sits around the action layer of agentic systems, where proposed actions must be evaluated, controlled, and evidenced before execution.
The question is no longer "did the AI system work?" — it is "can the institution prove that this specific AI-assisted action was authorised when it happened?" Start with a diagnostic: a structured assessment of where your agentic AI control and evidence gap is widest.