United States · Machine Authority Pilot
AI can bind your business.
Audact sits between agent intent and consequential effect. It evaluates configured authority before a governed action is released, and preserves evidence of what followed.
Synthetic demonstration · the model, not a running system
- Principal
- Acme Corp.
- Mandate
- Commercial Concessions · M-US-204
- Mandate ceiling
- $500,000
- Reserved
- $300,000
- Available
- $200,000
Agent proposes a $250,000 settlement
ESCALATE
Control is in development and is the subject of the pilot. The compliance receipt works today.
The consequentiality test
What can your AI actually cause?
Two questions decide whether this is your problem. What is your AI allowed to promise, and how do you know your business actually honored it? Audact begins where a machine can change something real. Pick what your agent can do today.
Synthetic demonstration · DEMO-US-001
One action. One explicit boundary.
A commercial-concessions mandate. The figures are chosen to show the mechanism end to end.
- Principal
- Acme Corp.
- Mandate
- Commercial Concessions · M-US-204
- Mandate ceiling
- $500,000
- Reserved
- $300,000
- Available
- $200,000
Agent proposes a $250,000 settlement
ESCALATE
The proposed commitment exceeds current available authority by $50,000. A person decides; both branches leave a receipt.
Agent proposes a $180,000 credit
Within available authority, within the per-action ceiling.
- Clearance
- Effect
- Obligation open
- Closure
- Authority released
The task is finished in seconds. The obligation stays open until something proves it was met, and the something is not the agent. The same mechanism carries live calls in Voice today.
The economics
From per-action approval to bounded autonomy.
The real competitor is not another vendor. It is a person approving every meaningful action.
Without an explicit boundary
- Every consequential action → a person
- Autonomy grows by trust, not by control
With an explicit authority boundary
- Within mandate → machine
- Outside mandate → a person
- Both branches leave evidence
We do not publish outcome percentages we have not measured. You already have the two numbers that matter: how many of these actions run per month, and what one approval costs you in cycle time.
The pilot path
Start without replacing your stack.
1 — Map
Identify one action class with real consequence: credits, spend, commitments.
2 — Observe
Establish current decision patterns and limits before anything is enforced.
3 — Control
Put that one action class behind an explicit, customer-defined authority boundary.
4 — Expand
Increase machine autonomy only where the evidence supports it.
Your organization defines the principal, the mandate, the ceilings and the approvers: your delegation of authority, made machine-enforceable. Audact enforces, accounts, preserves state and proves.
Two ways in
Who this is for.
I deploy AI inside my company
Control consequential actions inside enterprise workflows: finance, customer operations, procurement.
I build agents for clients
Add a consequence-control layer without rebuilding enterprise controls for every customer. You build the intelligence. Audact controls the authority.
Bring your stack
What Audact is — and is not.
Audact is designed to govern the consequence boundary, not to replace your model, your identity stack or your systems of record. The platform that carries the call belongs on that list too: Retell, Vapi or whatever you already run stays where it is. Bring the mandate. Audact governs the consequence. The mechanism sits on a patent-pending governance stack filed at the UK IPO.
- Your agent — Claude · GPT · Gemini · custom
- Your identity and access stack
- Audact — authority · clearance · consequence · proof
- Your systems of effect — CRM · ERP · payments · voice
Not an agent platform
Keep your agents.
Not IAM
Keep your identity and access stack.
Not a payment rail
Keep your payment infrastructure.
Not a legal decision engine
Your organization defines its mandate and its authority.
Proof without overclaiming
What a receipt proves. And what it does not.
Every governed action leaves a receipt. Precision is the differentiator, so the scope is explicit: on the page, not in a footnote.
What this proves
- The mandate reference and the authority input at decision time
- The requested value and the decision state
- That the receipt has not been altered since it was written
What this does not prove
- Who wrote the receipt: the integrity code is symmetric, not a signature
- Legal enforceability or external fulfillment
- Payment settlement or economic finality
Where this stands
Precision about the product, not marketing about it.
Every capability on this site carries a status word, and the same registry feeds every page. Nothing here is described as further along than it is.
Control
In development
The authority model, the clearance decision and the mandate state. It is the subject of the U.S. pilot.
Compliance receipt
One receipt per governed interaction. The sample files are open to anyone, without an account.
Voice
In controlled testing
Where Proof is already applied. Live calls run through it today.
The U.S. program is a paid 90-day pilot: one action class, one principal, one explicit boundary. Deliberately narrow, because a pilot someone pays for is the only kind that proves anything.
Start with one consequential action.
No platform migration, no replacement of your agent stack. Define one action class, one principal and one explicit authority boundary, and observe before anything goes live.
One conversation, with the people who built it.