AI in compliance, without exposing your clients or a live suspicion
Preparing the risk assessment, updating the policy, handling an internal report, yes, but never by pasting a client file or an alert into an AI without safeguards.
Risk assessments, internal policies, reports to management, internal reports, investigations: your files mix client identities, beneficial owners, alerts and suspicions. Pasted into a consumer AI without a data-processing agreement, they leave your control; banking secrecy and data protection law still hold you responsible.
Your colleagues and compliance officers already use AI to write. The point is not to ban it, but to set what goes into the tool and what never does.
- A risk assessment redone in a rush, which the board adopts without fully understanding it and the auditor finds incomplete
- A policy updated only in part, with a gap flagged by the audit firm
- Client files, alerts and suspicions pasted into public AI tools, outside any control
- An internal report or incident handled without a written procedure, to the detriment of the person who spoke up and the person concerned
See for yourself
A prompt structured in 5 blocks: Instruction · Role and audience · Data · Constraints · Format. An unframed AI invents; the frame stops it.
In a workshop, we adapt these prompts to your institution, each with its safeguards: no client data and no live suspicion goes into a public AI.
Risk assessment · annual
The head of compliance at a bank prepares this year’s AML risk assessment, or redoes it because the business has changed (new clients, new country, new product). They have the client statistics, last year’s assessment and the audit firm’s report to hand. This prompt produces a structured assessment that the board can adopt and the auditor finds complete: risk levels, measures, resources requested.
AML policy · update
The head of compliance at a bank must update the AML policy because the law has changed or because the audit firm has flagged a gap. They have the current policy, the audit remarks and the amended text of the law or ordinance. This prompt produces a policy that covers all the minimum content required, a table of changes the board can read, and the route to adoption by the right body.
Management report · compliance
At the end of the quarter or at the board’s request, a bank’s head of compliance prepares the report to management: alerts handled, reports filed, heightened risk relationships, findings, resource needs. They have the statistics, the internal control plan and the last report. This prompt produces a report that management reads quickly and that asks it for precise decisions, with a short version for the board.
Internal report · handling
An employee comes to see the designated officer: a colleague is said to push through transactions without clarifying them, or to accept gifts from a client. The officer has no written procedure for what follows and does not know whom to inform, how to protect the person who speaks up, or when a report to the financial intelligence unit arises. This prompt delivers the handling protocol and the acknowledgement of receipt that reassures without promising anything false.
Internal incident · investigation
Internal control finds that a client adviser opened relationships without having the beneficial owner declaration completed, or closed an alert without documenting it, and perhaps tipped off a client. Management asks for facts, not an impression, and the head of compliance must investigate without harming the employee or a file followed by the financial intelligence unit. This prompt delivers the investigation plan, the report, the proportionate sanction proposal and the data protection rules.
30 use cases cover your line of work in the Chaptal catalog, out of 750+ in the corpus · each one states what it produces, what it expects from you, and what it refuses to do.
“The Augmented Employee”
You and your compliance team learn to frame AI on your real files, assessments, policies, reports, without exposing a client or a suspicion.
See the training →AI usage audit
A review of your usage, a register and clear rules before a client file goes into the wrong tool.
See audit & advice →Do it in a conversation
Jacquard is the Palambur AI assistant. You describe what you need, it knows your line of work and the laws of your country, asks for what is missing and delivers an answer or a ready-to-use document. You choose the model for every question.
Frequently asked questions
Can I paste a client file or an alert into an AI to analyze it?
Not as it is. Without a data-processing agreement, what you paste leaves your control. You anonymize first, or you put nothing in; every card in the catalog repeats this safeguard.
Can AI decide that a suspicion must be reported to the authorities?
No, never. It can structure the facts and the documentation, but the decision to report a suspicion stays with the compliance officer, on their own judgment and under their own responsibility.
Does the AI know the law and the rules that apply to my institution?
It structures the approach and asks the right questions, but you hold the text. The AI helps you apply it, the decision stays yours, and the text in force prevails.
Is banking secrecy affected?
Yes. What you write into an AI tool can engage your professional secrecy. One card in the catalog covers what you may write about a client in such a tool, and what you must never write.
Do I need a large compliance department for this to be worthwhile?
No. A small institution or a wealth manager benefits first, because assessment, policy and reporting rest on few shoulders. The catalog prompts are used as they are, with no technical setup.