Research Summary: Governing Agentic AI in FinTech

Financial institutions are increasingly delegating critical decision-making to agentic AI systems, which operate with minimal oversight. This report highlights a significant gap in the governance of such AI, particularly within the FinTech sector, termed the 'Verifiability Gap.' This gap represents the shortfall between the required verification for delegated authority and the actual explainability and reproducibility of AI decisions. A new multilevel governance theory for agentic AI is proposed and partially tested, suggesting that the primary governance challenge is verifiability rather than AI capabilities.

Why it matters

The increasing reliance on autonomous AI in critical financial operations poses substantial governance and risk management challenges. Understanding and addressing the 'Verifiability Gap' is crucial for maintaining trust, ensuring regulatory compliance, and mitigating potential financial and reputational risks associated with AI-driven decisions. Establishing robust governance frameworks for agentic AI is essential for the responsible and effective integration of these technologies into the financial ecosystem.

Key insights

• Financial institutions are delegating consequential decisions to agentic AI systems that operate with limited human oversight.
• The governance of agentic AI in FinTech is currently under-researched.
• The primary constraint in governing agentic AI is identified as verifiability, not the system's capabilities.
• The 'Verifiability Gap' is defined as the discrepancy between the verification needed for delegated authority and the explainability/reproducibility of AI decisions.
• This gap is indexed to specific verifiers, evidentiary standards, and audit lag.
• A multilevel governance theory for agentic AI has been developed and is being tested through studies involving various AI model versions, from local to frontier systems.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.11344