AI agent training for risk managers in retail banking in Germany

German retail bank risk managers need training to judge whether an agent's authority is justified and whether its controls hold. Credit scoring of individuals is high-risk under the EU AI Act, DORA treats the agent platform as an ICT third party, and GDPR limits solely automated decisions. Training should make those links concrete. Risk appetite and credit decisions outside policy stay human.

What does a risk manager at a German retail bank do?

Germany's retail banking market is split between savings banks, cooperative banks and private banks, many of them small and regional, sharing central IT and service providers within their groups. That structure puts a lot of risk into outsourcing chains: the bank owns its risks, but much of the technology sits with a group provider.

Risk managers run the bank's risk inventory and stress testing, monitor credit portfolios such as overdraft lines, consumer loans and mortgages, oversee operational and ICT risk, and report to the management board and supervisory board. BaFin's supervisory requirements for risk management set detailed expectations, and larger institutions also answer to ECB banking supervision.

Where do agents change a German risk manager's work?

• Credit pre-assessment, depth 2. In the first line, an agent gathers income and account data, applies the bank's credit policy and drafts a recommendation on a consumer loan or overdraft increase. A credit officer decides. Because this touches creditworthiness assessment of natural persons, the risk manager must treat it as a high-risk use.
• Overdraft limit monitoring, depth 1 to 3. The agent watches account behaviour and drafts limit reviews. Reducing a limit is held, because the customer feels it immediately.
• ICT incident drafting, depth 2. The agent classifies ICT incidents against the bank's DORA criteria and drafts the internal report. A person confirms classification before any external notification.
• Outsourcing register review, depth 1. The agent reads contracts and service reports from the group provider and flags gaps in the bank's register of ICT arrangements.

Changing a credit limit, releasing a held payment or writing off a balance is depth 4 and held for a named approver.

What must risk managers learn, and what stays with them?

Three skills matter most, and they line up with the ones in which AI agent skills matter most for risk managers.

Reading certification as evidence. An agent's depth is earned by replaying it against the bank's own cases; the role sets a ceiling and the evidence sets the depth. A risk manager should check whether the replay included the bank's own customer mix, not just a group-wide sample, because a regional savings bank's borrowers differ from a city private bank's.

Writing approval rules. A held credit limit change waits for a named person. The rule that decides who that person is, and up to what amount, should be written by the risk function. Our guide to setting an approver rule for the risk team shows how.

Understanding withheld actions. Some actions are withheld at the platform floor for every agent at every depth. Risk managers should know the difference between that and the bank's own limits, which refuse or hold actions the bank has chosen to narrow.

What stays with people:

• credit decisions outside policy and any adverse credit decision a customer contests;
• risk appetite and stress-test conclusions;
• classification of major ICT incidents;
• reporting to the boards, BaFin and the ECB.

Which German and EU rules shape oversight and literacy?

The EU AI Act applies directly. Annex III lists creditworthiness assessment and credit scoring of natural persons as high-risk. Deployers of high-risk systems must follow Article 26, including assigning human oversight to people with the necessary competence, training and authority, and Article 14 sets what human oversight must achieve. Article 12 requires automatic logging. Article 4 requires deployers to ensure AI literacy among staff, which covers every risk manager, credit officer and approver working with agents.

GDPR applies to customer data. Article 22 restricts decisions based solely on automated processing that have legal or similarly significant effects, which a credit refusal plainly has. An agent should draft; a person decides. The German data protection authorities enforce GDPR at state and federal level.

DORA applies from 17 January 2025 and requires ICT risk management, incident reporting, resilience testing and oversight of ICT third parties. BaFin supervises German banks, with the ECB directly supervising significant institutions and the EBA setting common standards. Model risk is supervised in most banking markets, and Germany is no exception.

Where agents monitor staff activity, employee representatives often have co-determination rights over the tool. Involve the works council early.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.

A 90-day training plan

How risk managers fit alongside credit officers and ICT staff is mapped in AI agent skills by role.

Where the Academy fits

Pryme Intelligence Academy gives a German bank documented AI literacy across the risk function. Foundations is free; approvers earn Certified Pryme Operator through an exam and a two-week practicum. Risk leaders who oversee regulated, multi-agent deployments can take the Architect track, valid for 24 months, which ends in a board review with two Pryme assessors.