Generative AI has fast become a major talking point across the financial sector, but as financial firms shift from small pilots to large-scale deployment, the spotlight is on how this powerful technology can be embedded responsibly into anti-financial crime frameworks.
In a recent “Hype vs. Reality” fireside discussion, organised by World Salon alongside Columbia University’s Global Dialogue, SymphonyAI’s financial crime and sanctions specialist, Elizabeth Callan, tackled this very question. The episode, titled “What LLMs really mean for the future of finance”, brought together sector experts to unpack the real-world impacts of large language models (LLMs) on risk, compliance and regulation.
Drawing on over two decades of experience in financial crime policy, Callan outlined practical steps for weaving generative AI into operations while keeping regulators and stakeholders reassured.
A strong governance foundation is the first building block. While many firms now have AI risk committees in place, this alone is not enough. Clear accountability structures, detailed documentation and well-defined responsibilities are non-negotiable. “Most institutions now have AI risk committees,” Callan said, adding, “It sounds basic, but documented policies and procedures are absolutely critical.”
For both regulators and internal teams, transparency is vital. Organisations should maintain written policies that explain how AI models are governed, trained, tested and audited. Robust documentation of data sources, explainability standards and incident plans boosts trust and protects organisations as new AI regulation evolves.
Active engagement with regulators is another critical pillar. In the US, federal-level regulation is still lacking, though states have started to introduce laws around AI bias. Companies that build proactive dialogue with regulators will have more say in how frameworks develop — and will be better prepared to adapt, Callan explained.
Data is both the foundation and the risk for AI-led compliance systems. Organisations must ensure data quality and test continuously for potential bias. New fairness rules make it increasingly important for firms to prove that their AI models are effective, unbiased and fit for purpose.
Explainable AI is now a must-have, not a ‘nice-to-have’. Financial institutions need to be ready to demonstrate what data feeds a model, how it produces outputs, and how its decisions can be verified. “If you can’t explain it to a regulator, you probably shouldn’t be using it,” Callan said.
Human oversight remains essential. Combining human judgement with AI tools allows institutions to catch unexpected issues early. Pilot schemes run in sandbox settings help test new solutions in a safe space before rolling them out more widely.
Finally, strong vendor relationships can make or break responsible AI use. Firms should demand full transparency and long-term commitment from AI providers. Vendors must act as partners, helping clients refine strategies, maintain compliance and communicate clearly with stakeholders.
Read the story here.
Read the daily FinTech news
Copyright © 2025 FinTech Global
