Trustworthy AI in the financial sector: efficiency gains without loss of control
Artificial intelligence is changing the financial industry - and with the development of so-called "agentic AI", the next major upheaval is imminent. Agentic AI systems can make decisions and carry out actions independently - without every step being initiated by a human.
The potential is enormous. Especially in the highly regulated environment of anti-money laundering (AML) and anti-financial crime (AFC), however, this new autonomy brings with it a central challenge: how do financial institutions ensure that efficiency gains do not turn into a loss of control?
Where do we stand today? AI in the AML/KYC area
With our yourKYCplus solution, we experience every day how AI can speed up the onboarding process for new customers and the regular review of existing customers.
- Automated document interpretation: commercial register excerpts, shareholder lists and powers of attorney are digitally scanned, analyzed and structured with the help of AI.
- Faster compliance workflows: Information flows seamlessly into the audit and assessment.
This is already an important step in productive use - AI supports, humans decide. But the next evolutionary stage is just around the corner: agentic AI that not only provides suggestions, but also makes decisions and initiates processes independently.
The vision sounds tempting:
- Massive acceleration of processes and decisions in seconds in 24 X 7 mode
- Relieve compliance teams and focus on really complex cases.
- Cost efficiency through fewer manual activities and friction points in onboarding.
At the same time, however, new risks are emerging:
- Loss of control: Who is responsible when AI models make the wrong decisions?
- Lack of explainability: Can we explain to BaFin, the Financial Intelligence Unit (FIU) or auditors in a comprehensible manner why the AI made a certain decision?
- Regulatory uncertainty: The EU AI Act[1], BaFin circulars[2] and guidelines from the Financial Action Task Force (FATF)[3] are designed for traditional AI - not autonomous systems.
Managing the trade-off: our approach
Clear principles and concrete solutions are needed to ensure that agent-based AI can be used with confidence in the AML and AFC context. Our vision for yourKYCplus is based on five pillars:
- Human-in-the-loop: Even if AI works autonomously, humans remain involved and responsible for sensitive decisions.
- AI governance according to recognized standards: We are guided by the Fraunhofer IAIS guidelines[4] "Designing trustworthy AI" and the OECD Framework for the Classification of AI Systems[5]. This means: transparency, fairness, security and comprehensive documentation.
- Auditable decision-making processes (Explainable AI): All AI decisions are logged for internal quality audits as well as for regulators.
- Risk-based use: agent-based AI only takes decisions where the risk is manageable. Highly sensitive cases remain with humans for the time being.
Why act now?
The EU AI Act will be binding for financial institutions from 2026. It requires clear risk management for AI systems - especially for high-risk applications such as AML software.
- International organizations such as the FATF and the Basel Committee on Banking Supervision are already demanding that the use of AI must be traceable, verifiable and accountable.
- With yourKYCplus, we show that it is possible to combine innovation and compliance - and that the introduction of agent-based AI can be managed responsibly.
Denis Beau's quote sums up the central challenge: only clear rules create trust - and only with trust can innovation in the financial sector be sustainable in the long term. This is precisely why the controlled introduction of agent-based AI in AML and KYC processes is so crucial.
Conclusion: efficiency yes, loss of control no
is no longer a dream of the future. It is increasingly becoming the focus of compliance departments.
Financial institutions, ARCs and the central office are faced with a crucial question:
"How can the potential of AI be exploited without losing control over decision-making processes?"
yourKYCplus shows: The path does not lie in "all or nothing", but in a controlled transition - with clear guard rails, comprehensible decisions and the right amount of human supervision.
This is how we are creating a new generation of AML and AFC solutions: efficient, innovative - and trustworthy.
[1] EUR-Lex - 52021PC0206 - EN - EUR-Lex
[2] BaFin - Big Data & Artificial Intelligence
[4] AI test catalog - Fraunhofer IAIS
[5] OECD Framework for the Classification of AI systems | OECD
