SentriAI is the security inside Fintra. Agent Fence decides whether an action may run before it runs, the function that does it checks that decision first, and every decision is sealed into a hash-chained record an auditor can recompute.

Can an AI agent pay a vendor on its own?

No. In Fintra a person starts every vendor payment, and no AI agent has a tool that pays. An AI agent can ask Agent Fence for a decision, and the function that records the payment or writes the ACH file still redeems a one-use authorization bound to that exact payment before it runs, so an approval cannot be reused for a different payment, and one that has expired is refused. Fintra does not send the money: the payment leaves from your bank.

Does Fintra send the payment?

No. Fintra records the payment and can write the ACH file your bank sends; its own payment rail runs against a sandbox and moves no money. Where a payment reaches that rail, SentriAI keeps what Agent Fence authorized beside what the rail replied, and a payment it has not confirmed stays pending or unknown, never success.

No. Fintra holds no security certification. Its controls and evidence are mapped to the frameworks listed on the frameworks page, as evidence relevant to each, and the evidence is shared on request.

Does SentriAI replace our GRC tool or our SIEM?

No. SentriAI works beside them: it decides and enforces inside Fintra, and keeps the evidence your GRC tool, your SIEM and your auditor read.

SentriAI, the security inside Fintra

SentriAI decides before an action runs, and keeps a record an auditor can recompute.

Before a bill payment is recorded, an ACH payment file is written, a vendor's bank details change or a person's role changes, Agent Fence decides whether it may happen, and the function that does it will not run without that decision. Every decision is sealed into a record an auditor can recompute. Fintra moves no money: your bank sends the payment.

Book a demo Log in

How a payment passes SentriAI

1Ask. A person proposes a vendor payment, or an AI agent asks Agent Fence about one.
2Decide. Agent Fence allows it, holds it for a named approver, or denies it, and says why.
3Enforce. The function that records the payment or writes it into an ACH file redeems a one-use authorization bound to that exact payment before it runs.
4Pay. Your bank sends the money. Fintra’s own payment rail runs against a sandbox and moves none.
5Prove. The decision is sealed into a hash-chained record an auditor can recompute.

One security layer around payments, vendor bank details and who can act.

Decide before an action runs, keep what was authorized beside what happened, test the controls against attacks, and keep a hash-chained record an auditor can recompute.

Agent Fence

Decides before anything runs

Before a bill payment is recorded, an ACH payment file is written, a vendor’s bank details change or a person’s role changes, a person or an AI agent asks Agent Fence. One deterministic policy answers allow, hold for a named approver, or deny, with the reason written down. Anything it cannot establish is a deny.

Deny by default: an unknown action, a missing fact or an error is a deny
The actor, the company and the role come from the signed-in identity, never from the request
A second, independent person approves a payment above your approval threshold and every vendor bank change and role change, and the person who asked cannot approve their own request
An approval expires, and a used one cannot be presented again

Outcome record

Keeps what was authorized beside what happened

SentriAI does not take the word of whoever acted. When a payment reaches Fintra’s payment rail, which runs against a sandbox because Fintra moves no money, SentriAI writes down what Agent Fence authorized (the amount, the destination and the obligation) beside what the rail replied, and treats the reply as a claim, not a result. A payment counts as confirmed only when the rail’s own record of that exact payment is read back and matches; until then it is pending or unknown, never success.

Confirmed, pending, failed, reversed or unknown: the states SentriAI keeps for a payment
An unknown outcome is never recorded as success
The rail’s reply is kept as a claim beside what was authorized

Security Arena

How SentriAI is tested against attacks

SentriAI is tested with versioned adversarial scenarios, run offline against its real controls: unauthorized payments, false attribution, false completion and cross-company access. An independent oracle grades each run from the world’s own records, never from the answer of the control under test, and every result states what it was measured against.

Versioned scenarios and datasets
An oracle that imports nothing from the product it grades
Every result reported with its denominator

Saguaro

An investigation engine, run separately

Saguaro is SentriAI’s investigation engine. It runs as a separate service that Fintra does not host, and it is not connected to any company’s books. It reads security telemetry after the fact, never sits in the path of an action, and never acts without a named person.

A separate service, not hosted by Fintra
Not connected to any company’s books
Never acts without a named person

Evidence

A record an auditor can recompute

Each decision is sealed into a hash-chained record with its reasons, mapped to the controls it bears on. Fintra recomputes the chain and names any entry whose hash no longer matches. An administrator can also seal a signed checkpoint; kept outside the database, it shows a rewrite that recomputed the chain.

Hash-chained, and the database refuses updates and deletes
Mapped to the controls a framework asks for, as evidence relevant to it
Recomputed on request, entry by entry

The function itself checks the decision.

A screen can be skipped and an API can be called directly, so the check does not live in front of the work. Each of these functions redeems a one-use, signed authorization bound to that exact action before it writes anything, and refuses to run without one, whoever calls it.

Recording a bill payment
Generating an ACH payment file
Changing a vendor’s bank details
Adding a vendor’s bank details
Changing a person’s role
Approving a contractor’s onboarding
Handing an ACH payment to the payment rail
Deleting a company member or a payout vendor
Changing an approval threshold

Evidence relevant to the frameworks you answer to.

Each decision and each control test is mapped to the controls a framework asks for. The list says, for each framework, which controls Fintra carries and which it tests automatically.

See the compliance frameworks

Finance: SOX and PCI DSS, and a starter set for SOC 1, with client trust accounting, and Fintra's own sets for money held for others and for billing.
Security: SOC 2 and GDPR, and starter sets for ISO 27001 and HIPAA, with Fintra's own set for authority delegated to people and AI agents.
Customer success: Contact center controls, each tied to a SOC 2, GDPR or PCI DSS requirement; a caller is identified before anything is disclosed or changed.

What SentriAI is not.

1.Fintra holds no security certification. Its controls and evidence are mapped to the frameworks listed on the frameworks page, as evidence relevant to each, and the evidence is shared on request.
2.SentriAI works beside your GRC tool, your SIEM and your auditor, not instead of them.
3.An AI agent outside Fintra can ask SentriAI for a decision with its own agent key. Fintra enforces decisions in the functions listed on this page when Fintra runs them; an action the agent carries out in its own systems is the agent’s to stop.
4.Fintra moves no money. Its own payment rail runs against a sandbox, and your bank sends the payment.

Questions a security review asks

What is SentriAI?: SentriAI is the security inside Fintra. Agent Fence decides whether an action may run before it runs, the function that does it checks that decision first, and every decision is sealed into a hash-chained record an auditor can recompute.
Can an AI agent pay a vendor on its own?: No. In Fintra a person starts every vendor payment, and no AI agent has a tool that pays. An AI agent can ask Agent Fence for a decision, and the function that records the payment or writes the ACH file still redeems a one-use authorization bound to that exact payment before it runs, so an approval cannot be reused for a different payment, and one that has expired is refused. Fintra does not send the money: the payment leaves from your bank.
Does Fintra send the payment?: No. Fintra records the payment and can write the ACH file your bank sends; its own payment rail runs against a sandbox and moves no money. Where a payment reaches that rail, SentriAI keeps what Agent Fence authorized beside what the rail replied, and a payment it has not confirmed stays pending or unknown, never success.
Is Fintra certified?: No. Fintra holds no security certification. Its controls and evidence are mapped to the frameworks listed on the frameworks page, as evidence relevant to each, and the evidence is shared on request.
Does SentriAI replace our GRC tool or our SIEM?: No. SentriAI works beside them: it decides and enforces inside Fintra, and keeps the evidence your GRC tool, your SIEM and your auditor read.

See SentriAI stop a payment it should not make.

Book a 15-minute demo: we will follow one vendor payment from an agent's request to Agent Fence's decision and the approval, on a simulated bank rail with sample data.