Banks are struggling to scale AI safely as regulatory expectations fragment globally. Ratul Ahmed highlights the need for stronger governance, clearer definitions, and human accountability to balance innovation with compliance.
Financial
institutions are accelerating their adoption of artificial intelligence, but
many are finding that the real challenge lies not in innovation, but in safely
scaling these technologies under increasing regulatory pressure.
Ratul Ahmed, Group Head of Model Risk
Management and Validation at Commerzbank, said the industry is grappling with a
complex balancing act between speed, trust, and compliance as AI becomes
embedded across a widening range of business processes.
“Financial institutions are racing to
use AI to drive efficiency and innovation, but our immediate struggle is how to
scale safely and quickly under tightening and fragmented rules,” Ahmed said.
She emphasized that AI is no longer confined to traditional risk and pricing
models, but is now permeating decision-making across the enterprise, increasing
both opportunity and risk.
One of the most pressing challenges
is governance. As firms deploy AI at scale, existing frameworks are being
tested in ways they were not designed to handle. Ahmed noted that many
institutions are revisiting their three lines of defense structures to assess
whether they remain fit for purpose in an environment defined by rapid
operational velocity.
“In several ways, firms are
adapting,” she said, pointing to measures such as risk-based tiering,
enterprise gating, and clearer internal definitions of AI systems and use
cases.
These approaches are intended to
ensure that controls scale alongside technology, rather than lag behind it.
A recurring issue is the lack of
clarity around what constitutes an AI model versus an AI system. Ahmed warned
that conflating the two can create confusion in governance and lead to gaps in
oversight.
“Providing a clear definition of AI
through the organization is critical,” she said, adding that firms must be able
to differentiate between models, systems, and broader applications.
Regulatory expectations are also
evolving, but not uniformly. Ahmed highlighted the growing fragmentation across
jurisdictions, which is making compliance more complex for global institutions.
“Regulatory expectations differ
across geography,” she said, noting that firms often misjudge the level of
scrutiny applied to explainability, particularly in areas such as credit and
capital decision-making.
Explainability remains a central
concern. While AI can enhance predictive capabilities, it can also obscure how
decisions are made if not properly governed. Ahmed cautioned that many firms
underestimate the importance of transparency, particularly when regulatory
approval is required.
