Sebastian Nordgren, Global Head, Privacy & AI Governance at MUFG, explains why weak data governance threatens financial institutions' AI ambitions. He examines how established risk frameworks can support responsible innovation, why autonomous agents require stronger safeguards and how trustworthy, explainable AI could become a decisive competitive advantage.
As financial institutions accelerate artificial intelligence
adoption, weaknesses in traditional data governance threaten to undermine their
ability to manage increasingly sophisticated systems.
Sebastian Nordgren, Global Head, Privacy & AI Governance
at MUFG, argues that trustworthy AI depends on strengthening existing
governance foundations rather than creating entirely separate oversight
structures.
For many organizations, the challenge predates generative AI.
Data quality, cataloging and lineage have frequently been treated as isolated
projects rather than continuous governance responsibilities.
The emergence of autonomous AI systems is exposing the
limitations of that approach.
Nordgren identifies poor coordination between business
specialists and data governance teams as a persistent obstacle.
Differences in terminology, inconsistent quality expectations
and incomplete data mapping can undermine accountability before AI systems are
deployed.
The problem becomes more complex as institutions move beyond
structured information toward the unstructured, high-entropy data increasingly
used by advanced AI applications.
Addressing these weaknesses requires broader stakeholder
participation, revised operating models and sustained leadership commitment.
Building Governance Into Existing Controls
Rather than constructing entirely new governance
arrangements, Nordgren advocates extending established risk management,
privacy, cybersecurity and compliance frameworks.
This approach allows financial institutions to address
AI-specific vulnerabilities without unnecessarily slowing innovation. Dedicated
AI policies may still be required to satisfy regulatory expectations, but
existing controls should provide the starting point.
Third-party risk assessments, vendor requirements and
contractual provisions governing AI-related liability are among the areas
requiring attention.
Data provenance is equally important. Institutions must
understand where information originates, how it moves through systems and who
remains accountable for its use.
Nordgren highlights privacy impact assessments, data
protection assessments, retention policies and records management as
established mechanisms that can support AI oversight.
Cybersecurity capabilities, including identity management,
access controls and data loss prevention, must also evolve alongside
increasingly complex AI deployments.
Consistent information classification and recognized
frameworks, including the EU AI Act and NIST AI Risk Management Framework, can
help establish a common governance language.
Autonomous AI Demands Stronger Safeguards
The emergence of agentic AI introduces additional challenges
because systems can independently interact with applications, make decisions
and execute actions.
Nordgren recommends treating AI agents similarly to new
employees, applying least-privilege access principles and carefully
scrutinizing requests for elevated permissions.
Continuous monitoring of agent behavior, token consumption
and emerging risk indicators should accompany rigorous testing of operational
safeguards.
He also emphasizes the importance of thoroughly tested
emergency shutdown mechanisms and formal onboarding procedures for new data
sources and application programming interfaces.
These protections become particularly important when
autonomous systems can access sensitive information or interact with critical
business processes.