- Home
- Industries
- Banks and financial services
AI agents for banks and financial services
Banks, payment institutions and investment firms run on careful, repeated document work. We build agents for that work, keep them away from credit decisions unless your risk team builds to the high-risk standard with us, and contract as the ICT supplier DORA expects.
The problems we hear first
KYC files that stall
Analysts spend their day chasing the same missing documents and re-reading register extracts, not judging risk.
The same customer questions
Fees, limits, card blocks, statements: the answers are in your terms, but every channel asks again, in several languages.
Alert queues
Each monitoring alert needs the same file built by hand before anyone can judge it.
Supplier due diligence
Every new tool needs a register entry, an exit plan and audit rights. A vendor who can’t answer slows the project.
Each one prepares. A person decides.
Examples of what we build in this sector, not client work. For each one: where a person approves, and what it never does.
KYC file assistant
Checks the file against your checklist, lists what is missing or out of date, and drafts the request to the customer.
- A person approves
- An analyst accepts the file, asks for more or refuses.
- It never
- Accept or refuse a customer, or set a risk rating.
Customer service agent
Answers from your approved terms and FAQs, in the customer’s language, says it is an AI and hands over to a person.
- A person approves
- Complaints, advice and exceptions go to your team.
- It never
- Give investment or credit advice, or promise what your terms don’t.
AML alert preparation
Collects the transactions, documents and earlier cases behind an alert into one file, with the source of each item.
- A person approves
- An investigator closes or escalates the alert and decides on any report.
- It never
- Close an alert, contact the customer about it, or file a report.
Exceptions and reconciliation
Matches entries, explains each one it could not clear, and drafts the correction.
- A person approves
- An operations analyst signs off every correction.
- It never
- Post an entry or move money.
One job, step by step, with the person inside the flow.
AML alert file
A transaction-monitoring alert fires on a business account. The agent builds the file; an investigator decides.
- Sector
- Banks
- Risk class
- Not listed in Annex III; reasons kept
- Who decides
- An AML investigator
- Alert receivedIncoming transfers well above the account’s usual pattern: six in nine days (sample).Done
- Only the alert’s scope readThe account’s last 90 days, its KYC file and earlier alerts on this customer; nothing else.GDPR Art. 5(1)(c)Done
- File assembledTransactions, counterparties and the customer’s stated business, each with its source.Done
- Two points flaggedA counterparty outside the stated business; one invoice reference used twice.Flagged
- Neutral summary draftedFacts and sources only. No view on suspicion.Drafted
- Waiting for an investigatorThe agent never closes an alert, contacts the customer about it or files a report.Close with a reasonAsk for internal dataEscalateAI Act Art. 14GDPR Art. 22Waiting for a person
- Next step takenOnly after the investigator’s decision.Locked
- Run recordedEvery source opened, the draft, and who decided.AI Act Art. 12On record
A sample, scripted on invented data: no model is called and nothing here is client work.
Where the law draws the line for you
Each rule with the day we read it. An engineer’s reading for your counsel to confirm, not legal advice. How we read the AI Act.
| Rule | What it says (our summary; quotes are verbatim) | What it means for an agent |
|---|---|---|
| AI Act Annex III 5(b)Read 12 November 2026 | Lists AI used “to evaluate the creditworthiness of natural persons or establish their credit score”, except AI used to detect financial fraud. | An agent that scores or ranks credit applicants is high-risk from 2 December 2027. KYC files, alert preparation and customer questions are not in this point. |
| AI Act Art. 6(3), 6(4)Read 12 November 2026 | An Annex III system is not high-risk where it “does not pose a significant risk of harm”, for example a narrow procedural or preparatory task, but it always is when it profiles people. | We write the classification down with its reasons. A provider relying on the exception documents it and registers the system (Art. 49(2)). |
| DORA Art. 2(1), 28, 30Read 12 November 2026 | Credit, payment and e-money institutions and investment firms are in scope. They stay “fully responsible”, keep a register of information on ICT contracts and need exit strategies for critical functions; Art. 30 lists the contract terms. | Our contract carries the Art. 30 terms: the service, data locations, incident help, audit and access rights, termination and exit. |
| NIS2 Annex I, Art. 21(2)(d)Read 12 November 2026 | Banking and financial market infrastructures are sectors of high criticality, and entities manage supply-chain security with their direct suppliers. | We answer your supplier security review layer by layer, including what is still partial. |
| GDPR Art. 22Read 12 November 2026 | A person has the right not to be subject to “a decision based solely on automated processing, including profiling” with legal or similarly significant effects. | The agent prepares; a named person decides anything that affects a customer. |
| AI Act Art. 50(1)Read 12 November 2026 | Providers make sure people are told they are interacting with an AI system, unless that is obvious. Applies from 2 August 2026. | Customer-facing agents open with an AI notice in their first message. |
Nine layers around every agent, each with what is partial.
- Access and identityGoogle sign-in, revocable keys, an agent’s rights narrowed by its spec.Partial No MFA of our own and no enterprise SSO; open sign-up is on.
- Separation between clientsEvery record and search index filtered by the signed-in identity; a live agent gets its own containers and database.Partial Logical separation in one shared database, not yet tested independently.
- Hosting, network, encryptionHetzner, Nuremberg; HTTPS with HSTS; no inbound web ports; credentials sealed.Partial Stored data in general is not encrypted at rest.
- Only the data the task needsThe agent reaches only the systems its approved spec declares; secrets are scrubbed.Partial Personal data is not masked before the model.
- The model stepProvider set per deployment; each request admitted and recorded before it is sent.Partial The region is named in your contract, not promised EU-only.
- Untrusted content and testingDocuments are data, never instructions; a safety pack that must all pass.Partial No independent penetration test yet.
- A person decidesWrites wait for a person; a person approves every release; Stop and spend caps.Partial A live agent’s held actions are approved by our operator, not on your screen.
- Every run on recordEvery run recorded step by step, with approvals and who decided.Partial Append-only by convention, not tamper-evident; no client export screen.
- Deletion, export, backupsErasure with a receipt per step; a backup before every release.Partial Export and account deletion by hand, on request; backups not encrypted.
The supplier terms a supervisor expects.
If you are in DORA’s scope (Art. 2(1)), you stay fully responsible for your ICT risk, and your supplier contracts carry the Art. 30 terms. We sign them.
- Register of informationThe facts your register needs for our contract: the service, the functions it supports, our sub-processors and where data is processed (Art. 28(3), Art. 30(2)(a), (b)).
- Audit and access rightsFor critical or important functions, unrestricted rights of access, inspection and audit for you, an auditor you appoint and your supervisor (Art. 30(3)(e)).
- An exit planAn exit strategy with a transition period, and the return of your data in an easily accessible format (Art. 28(8), Art. 30(2)(d), 30(3)(f)). Export is by hand today, on request.
- Incident reporting supportHelp in an ICT incident related to our service at a cost agreed in advance, fast enough for your initial, intermediate and final reports (Art. 19, Art. 30(2)(f)).
A timeline up front, a fixed price after one call.
Most teams start with a readiness review of the process they have in mind, or go straight to a pilot when the use case is clearly outside the high-risk list. We don’t publish prices: the proposal after a short scoping call gives a fixed price for your scope, in EUR, excluding VAT.
All engagements and what moves the priceBook a callPrice on request
AI Act and GDPR readiness for one use case.
What’s in itBook a callPrice on request
One agent on one process, tested and approved.
What’s in itQuestions from banks and financial services
Is a KYC agent high-risk under the AI Act?
Not by itself. Annex III 5(b) covers the creditworthiness and credit scoring of people; checking a file against a checklist is not in that point. It still handles personal data, so GDPR applies in full, and a person makes the onboarding decision. We write the classification and its reasons down for your file.
Can you sign the DORA contract terms?
Yes. Our contract covers the Art. 30(2) terms for all ICT services and, for critical or important functions, the Art. 30(3) terms, including unrestricted access, inspection and audit rights and an exit plan with a transition period. We give you what your register of information needs.
Where is the data processed?
The platform runs in Nuremberg, Germany. The model step’s provider, route and region are named per deployment in your contract; we don’t promise that it stays in the EU unless that route does.
Do you connect to our core banking system?
Through its API, where it has one, built and tested in your project. Writes ask a person each time unless you decide otherwise in writing. Many first agents only read.
Related pages
Posts on these rules
Bring one process. Leave with a risk class, a pass bar and a plan.
A 45-minute scoping call with the people who would build it. You get a first view of the AI Act risk class, the controls it needs, and whether an agent is the right tool at all.