What the agent never does
The things the agent will not do, whatever it is asked.
Everything below is a constraint on the agent. An administrator cannot turn any of it off.
- Reads the transfer, the counterparty and the exposure
- Raises findings, each with its source attached
- Recommends an outcome, citing findings graded as evidence
- Drafts a report from your organisation's template
- Records a screen that did not answer as unavailable
- Decides the outcome
- Closes the case
- Submits the report
- Lodges it with the regulator, outside Cogentic
The boundary is a constraint, not a setting. An administrator cannot move an item across it and no permission level unlocks one.
Seven constraints, in one shape. None of them is a setting.
It never files with a regulator
The agent lodges nothing with AUSTRAC or any other regulator. Nothing else in Cogentic does either. Submitted is a state inside the product. Cogentic does not record whether you lodged a report, does not track lodgement deadlines and cannot tell you whether you met one.
It never closes a case
A case closes when a person selects one of the two resolving outcomes. The agent produces a recommendation. Nothing applies it until a person does.
Auto-assignment is routing, not a decision. A case being assigned to somebody says nothing about what should happen to it.
It never blocks, freezes or restricts by itself
The agent holds no control over a customer’s account. It cannot freeze funds, block a transfer, suspend an account or restrict a customer, and neither can anybody else from inside Cogentic. Those controls live in the systems that hold the funds.
What an investigation produces is a record and a recommendation. Acting on either happens in your own platform, under your own approvals, and Cogentic does not record whether you did.
It never uses a lead to justify an outcome
Findings are graded. The agent may cite a finding graded evidence as a reason for an outcome. It may not cite a lead. Cogentic strips any such citation before you see it.
It never treats a check that did not run as a check that passed
When no sanctions list answers, Cogentic records the screen as unavailable, not clean. The agent will not recommend closing the case as a false positive. It recommends holding or escalating, and it names the subject that went unchecked.
A vendor that failed, or was never configured, leaves no trace on the sanctions screen. You never see coverage you did not get.
Verification never clears onchain risk
A customer passing KYC can raise concern. It can never resolve it. A fully verified, long-standing customer does not explain away exposure to a sanctioned entity, and the agent still escalates.
It never invents an identifier or a citation
Every finding a recommendation cites is a real finding on the case. The agent cites it by its own identifier and shows it beside the reasoning, so you can open it. Where the agent holds only a partial identifier, it writes the partial one.
What this means when you sign
| The agent did this | A person did this |
|---|---|
| Consulted the sources and recorded what each returned | Chose which vendors are connected |
| Wrote the findings and graded them | Uploaded the documents a finding may cite |
| Proposed an outcome with its reasoning | Selected the outcome |
| Drafted the report from a template | Reviewed it, submitted it and lodged it |
You are accountable for the right-hand column. The case records the left-hand column with its sources, so you can check it.
Related
What Cogentic is
Cogentic investigates the alerts your systems already raise and drafts the report. It raises no alerts of its own and it lodges nothing.
Deciding a case
Only a case's assignee can record an outcome, and two of the five end the case for good.
The audit log
Cogentic records every change to a case against whoever made it. Reading one is not a change, so it leaves no entry.
Last checked against the product on .