The Financial Stability Board (FSB), a global financial watchdog, has published 12 sound practices for the responsible adoption of AI in financial institutions. The new guidance comes as agentic systems move deeper into regulated or higher-risk workflows, planning and executing tasks with increasing autonomy. This has made board-level oversight all the more important for banks, payment firms and other fintechs as they move beyond experimental AI tools into consequential production workflows.
Although the report offers guidance on the responsible adoption of AI in finance, it doesn’t amount to binding legislation. It instead reflects months of consultation and stakeholder feedback. The 12 practices cover organization-wide governance and risk management throughout the AI lifecycle, rather than deployment alone. Given that boards and senior managers are ultimately responsible for strategy, accountability and AI risk management, the publication provides a useful toolkit that organizations can adapt to their particular risk profiles.
The publication emphasizes stronger expectations for human oversight whenever an AI use case is deemed higher-risk or more autonomous—such as those that involve credit decisions or the movement of money. In such cases, approval triggers, escalation paths and strict limitations on delegated authority are essential when an agent can autonomously modify records or initiate actions. As Reuters describes, the guidance is a direct response to growing concerns about agents interpreting goals in unpredictable ways. That’s not to suggest every AI-generated recommendation or action needs to be individually signed off by a person, however.
Why autonomous AI needs stronger controls
Accountability for an AI agent’s unintended actions ultimately rests with people, not machines. The recommendations thus span model selection, data quality, explainability, evaluation, and continuous monitoring. Moreover, testing must be proportional to the materiality of a use case, rather than being treated purely as a pre-launch exercise. For example, if an AI agent can’t explain its actions adequately, it might instead be paired with compensating controls like increased human oversight, rather than being ruled out entirely.
Cybersecurity and third-party dependencies are also major focus areas of the FSB’s publication. For example, risk can increase due to AI models being sourced externally, opaque vendor claims, or high concentration on a relatively small number of providers. In such cases, service continuity and contractual transparency are essential, particularly in workflows that connect to regulated data, involve direct interaction with customers, or involve any regulated function.
What fintechs need to do next
Although AI oversight initiatives vary in scope and legal status, they all share a common principle: high-risk uses require demonstrable controls. When it comes to real use cases in financial services, such as fraud triage, credit decisions, or any other consequential workflow, these principles are essential both for long-term success and risk management.
In practical terms, this means financial services organizations and fintechs need to maintain appropriate audit trails of instructions, permissions, tool use, and intervention decisions. Nonetheless, controls should still be proportionate so that they protect customer trust without holding back genuinely beneficial automation. That’s ultimately what makes auditable AI controls an important differentiator in both enterprise procurement and partnerships between banks and fintechs.