AI That Does the Work - Behind an Approval Gate
AgentFence defines what each AI agent may touch, SentriAI records every action, and consequential steps like posting or paying always require a human to approve first.
Illustrative product view
What governed AI in Fintra does
The risk with AI in finance isn’t that it does the work - it’s that it does the work unsupervised. Once agents can categorize transactions, prepare filings, and schedule payments, the agents themselves become a control surface. Fintra governs them with two layers: AgentFence defines what each agent is allowed to do, and SentriAI records every action and produces the evidence an auditor wants. AI handles the toil; a human approves anything consequential before it happens.
- Per-agent policies: exactly what each AI may read, draft, and execute
- Mandatory approval gates before money moves, filings submit, or entries post
- A complete audit trail of every agent action - including blocked attempts
- SOC 2 evidence generated from the controls, not assembled the week before an audit
Core capabilities
| Capability | What it does | What it replaces |
|---|---|---|
| Agent permissions | Bounds what each agent may read, draft, and execute | Ungoverned automation scripts |
| Approval gates | Forces human sign-off before money, filings, or postings | AI acting with no checkpoint |
| Audit trail | Logs every agent action, allowed or blocked, with context | Log exports stitched across tools |
| Policy versioning | Shows which rules governed the agents at any point in time | “Which policy was live then?” guesswork |
| SOC 2 evidence | Exports control activity and approvals for the audit period | Pre-audit evidence scrambles |
How the approval gate works
From an agent’s action to an approved outcome
- 1
Set the policy
For each agent, define what it may do on its own, what needs approval, and what’s forbidden - in plain policy, not code.
- 2
Agent proposes
The agent does its work - coding an entry, preparing a filing, drafting a payment - and produces a proposal, not a final action.
- 3
Gate consequential steps
Anything that moves money, submits a filing, or posts to the ledger is held at an approval gate the agent cannot pass alone.
- 4
Human approves or rejects
A named person reviews the proposal with its context and approves or rejects it; only then does the action take effect.
- 5
Everything is logged
The proposal, the decision, and the policy version in force are written to the SentriAI audit trail as evidence.
AgentFence and SentriAI: two jobs, one layer
The two names do distinct work. AgentFence is the policy layer that decides what an agent is allowed to do before it acts - the gate. SentriAI is the control and audit layer that records what happened and turns it into compliance evidence - the ledger of trust. Together they make Fintra’s AI governable: bounded in advance, watched in the moment, and provable after the fact.
- AgentFence - per-agent permissions and mandatory approval gates, set before agents act
- SentriAI - the tamper-evident audit trail and control evidence across the platform
- Policy versioning ties every logged action to the rules in force when it happened
- The same layer governs finance, payroll, tax, and HR agents, not just one module
How it connects to the rest of Fintra
- Bill pay: agents draft and schedule, but release of any payment is gated to a human
- General ledger: AI categorizes, a person approves before entries post
- Sales tax and payroll: agents prepare figures; filings and payments need sign-off
- Compliance: control evidence and the audit trail feed straight into SOC 2 prep
Frequently asked questions
What is the difference between AgentFence and SentriAI?
AgentFence is the policy layer that decides what each AI agent may do before it acts, including the approval gates on consequential steps. SentriAI is the control and audit layer that records what actually happened and turns it into compliance evidence. AgentFence bounds the agents; SentriAI proves how they behaved.
Can an AI agent post entries or move money on its own?
No. Consequential actions - posting to the ledger, submitting a filing, releasing a payment - are held at an approval gate the agent cannot pass alone. Agents draft and propose; a named person reviews the proposal with its context and approves it before it takes effect, and both the proposal and the decision are logged.
How does governed AI help with a SOC 2 audit?
The evidence a SOC 2 audit needs - who approved what, which controls ran, what the AI was allowed to do - is produced continuously as agents work and humans approve. Instead of assembling artifacts the week before an audit, you export the period’s control activity, approvals, and policy versions from the audit trail.
What happens when an agent tries to do something it’s not allowed to?
AgentFence blocks the action and SentriAI records the attempt with the policy that forbade it. That blocked-attempt log is valuable evidence: it demonstrates the guardrail is enforced in practice, not just described on paper. Policy versioning means you can also show exactly which rules were in force at that moment.
Stay in the loop
One practical finance briefing a week - new guides, checklists, and benchmarks.
Let AI do the work, keep the control
Start free, no card required. Bound your AI agents with policy and approval gates today.
Talk to us