Fintra Feature

Let an agent run the pay run: inside a sealed mission

A human approves a mission envelope, which invoices, which vendors, which account, per-payment and aggregate caps, an expiry. Only then may an agent act. Every payment is reconciled back to that approved intent, and anything outside it is held before the wire.

Verified Autonomous Finance

Example figures
MISSION CAP
$412,000
38 approved invoices
RECONCILED
36 verified
match approved intent
HELD
2 lines
outside the envelope
Pay approved invoice INV-1041
verified
Invoice added after mission sealed
held
Vendor bank account changed pre-run
held
Aggregate within $412k cap
within mission
Assurance receipt (sha256) written
sealed
Illustrative product view

Autonomy is useful; unbounded autonomy is a liability

The moment an AI agent can initiate a payment, the question stops being “is the model good?” and becomes “should this specific transfer happen, right now, given what a human actually approved?” Verified Autonomous Finance answers that question by putting a sealed boundary around the agent before it can move a dollar, and by proving after the fact that it stayed inside the lines.

Intent versus state: reconciled line by line

OutcomeMeaning
verifiedMatches an approved invoice, vendor, and the caps
permitted exceptionDiffers, but inside a pre-authorized tolerance
heldOutside the envelope: added late, over cap, vendor changed
unauthorized varianceA payment with no approved basis at all
unable to verifyEvidence missing: never silently passed
Every payment lands in one of five honest states

Every run ends in a sealed receipt

From mission to assurance receipt

  1. Human seals the mission

    Approved invoices, vendors, source account, caps, prohibitions, and a TTL are hashed into a Mission Envelope.

  2. Agent proposes payments

    Each proposed payment is checked against the sealed envelope and the running aggregate.

  3. Intent is reconciled to state

    reconcile_intent_to_state diffs what was approved against what actually happened, labelling every line.

  4. An assurance receipt is written

    The whole run is sealed with a canonical sha256 hash so the outcome is provable and tamper-evident.

Questions people ask

What is Verified Autonomous Finance?

It is a control that lets an AI agent execute a pay run only inside a human-approved, cryptographically sealed mission, then reconciles every payment back to that approved intent and holds anything that drifts outside it.

Does it actually move money?

Not on its own. It runs in shadow mode by default: it decides and proves what should have been held or blocked without disbursing. You enable enforcement when you trust the record.

What happens if an agent tries to pay something that was not approved?

That line is labelled held or unauthorized variance and drops out of the run with a reason. A payment with no approved basis is never treated as verified.

How is the outcome provable to an auditor?

Every run ends in an assurance receipt sealed with a canonical sha256 hash over the decision, and each decision is also written to the tenant-scoped Decision Ledger.

Keep reading

One finance briefing a week

New guides, checklists and benchmarks, by email. Unsubscribe from any issue.

 

Put a decision, with proof, in front of every action

See how Fintra decides allow / step-up / hold / block per action and writes each verdict to a tamper-evident ledger.