With volatility now a constant, the most effective risk leaders are those who pair disciplined thinking with curiosity and human judgment to anticipate disruption and navigate uncertainty when models fall short.
At CeFPro’s Risk Evolve 2025 conference event earlier this year, senior executives from global banks, asset managers, and venture investment firms warned that the financial sector’s rush to deploy artificial intelligence is being hampered by poor data quality, fragmented governance, and a ‘new toy’ mentality that risks running ahead of oversight.
A chief risk officer at a large European asset manager said the venture capital world illustrates both the promise and the limits of AI in risk management.
Traditional credit models, she explained, simply cannot assess startups that have no performance history or financial track record.
“We are experimenting with generative AI to enhance our due diligence and automate elements of KYC,” she said. “But human oversight remains essential, especially when judgment and context define the investment decision.”
A senior quant from a major global bank described AI’s use on the trading floor as more evolutionary than revolutionary.
“There’s huge potential for sentiment analysis and automation,” he said, “but AI systems still fail at basic logic because many depend on separate computational agents that don’t integrate well.”
His example - an AI model that could write fluent answers but couldn’t add two numbers correctly - drew laughter but emphasized a serious point: reliability and explainability remain major hurdles to adoption in financial markets.
For another expert, a senior board member at a universal bank, the key challenge is governance.
“Every institution is under pressure to demonstrate AI progress, but few have defined the right use cases or guardrails,” he said. “We need coherent frameworks to decide which tools are appropriate and how to oversee them without overregulating innovation.”
A senior operational risk executive from a multinational bank added that the enthusiasm for AI often outpaces maturity.
“Everyone wants to play with the new toy,” he said. “But we have to be the adults in the room asking: why, and what’s the return?”
He emphasized the need for education around model choice and control, warning that “there’s no one-size-fits-all approach” and that different systems carry very different data security risks.
He also cautioned that most organizations underestimate how poor data quality undermines their AI ambitions.
