Generative and agentic AI could transform risk and control self-assessments from periodic compliance exercises into dynamic risk management tools. But financial institutions must first improve processes, controls and data, with risk leaders warning that automating a weak framework will simply make its weaknesses operate faster.
Financial institutions hoping
artificial intelligence will transform risk and control self-assessments must
first confront a less comfortable question - whether the underlying process is
worth automating.
AI can multiply the effectiveness of
a strong framework, but applying powerful technology to inefficient processes,
weak controls or poor data will not suddenly create better risk management.
Speaking at CeFPro’s 2026 Risk Evolve
event, a director responsible for control intelligence and control monitoring
automation at a major UK bank described AI as a multiplier.
“If the process is muddled with
issues, then a multiplier into zero is zero,” she said.
That challenge goes to the heart of
the RCSA’s uncertain status within operational risk
A head of enterprise risk at a major
UK insurer said firms still need to understand their largest risks, whether
controls work and where investments or action plans are required. RCSAs support
those decisions, but their traditional form has significant weaknesses.
They can be time and people
intensive, point-in-time assessments and vulnerable to subjectivity or bias.
The objective should therefore be to retain their ability to identify exposures
while making the process more efficient and effective.
The banking executive was more
explicit about the perception problem. For first-line employees, she said,
RCSAs can feel like “a mundane operational task” or an annual exam administered
by the second line.
Yet the same framework could become
trigger-based and data-driven, with elements automated and ultimately capable
of supporting continuously updated risk management.
The determining factor is how mature
the institution’s existing approach is.
Technology itself is only part of the
challenge. A chief product officer at a risk technology provider warned that
implementations frequently encounter obstacles around data access, security,
certification, segregation of duties and internal permissions.
“There are many, many different
problems that are not technology based,” he said, identifying people and
processes as significant barriers. Institutions may be attracted by the
potential of agentic AI without understanding how to move from ambition to implementation.
The structure of the assessment also
matters. The banking executive argued that effective risk management requires
both bottom-up understanding of individual processes and controls and a
top-down view of principal risks.
Material differences between those
perspectives can reveal weaknesses in how risk is being understood or
communicated.
The insurance risk executive
highlighted another complication - overlapping exposures. Third-party, cyber
and AI risks increasingly intersect, raising the prospect of institutions
repeatedly assessing connected risks in isolation.
She suggested the direction of travel
should instead be toward understanding a “risk web.”
AI could help institutions navigate
that complexity, but only if deployed against clearly understood objectives.
Firms should begin with their
strategic direction, identify the processes supporting it and then examine the
associated risks and controls.
That principle also challenges
institutions to reconsider the value of RCSAs themselves. The technology
executive said larger organizations too often treat them primarily as
compliance exercises.
“I think you are leaving a lot on the
table,” he said, arguing that risk represents uncertainty rather than simply
threats and can therefore inform strategic opportunity. Properly constructed
assessments can become “an incredible asset for business” and strategic
decision-making.
The insurance executive agreed,
describing RCSAs within her organization as a critical decision-making tool
supporting discussions about investment in areas including cyber risk.
The opportunity, she argued, lies in
changing how assessments are conducted so they generate greater value.
For AI-enabled RCSAs, data quality
and explainability remain significant constraints. Asked about identifying
incorrect assumptions or bias in agentic systems, the technology executive
acknowledged that doing so was “not easy at all.”
Bias can be embedded within training
data, while organization-specific datasets may be limited.