QuiverCrypto QUIVERCRYPTO SUBSCRIBE
QuiverCrypto
← Blog

Shifting supervisory frameworks for agentic AI in finance

Explore the evolving supervisory architecture for agentic AI as outlined by the FSB's Sound Practices report to G20 finance leaders.

21 November 2026 · 5 min read
Shifting supervisory standards-but-still-gaps-remain/">frameworks for agentic AI in finance

The conversation around the regulatory landscape for agentic AI in finance is rapidly evolving. Recent developments signal a transition from mere theoretical discussions to actionable frameworks. The Financial Stability Board (FSB) has submitted its final report, presenting a set of sound practices to the G20 Finance Ministers and Central Bank Governors. This milestone was marked by their meeting in Bangkok, where they reinforced the message that unregulated experimentation with autonomous AI in finance is dwindling.

Although the guidelines provided by the FSB are non-binding, they establish essential standards for risk management. Organizations that disregard these guidelines may do so at their own risk, as highlighted in a letter from FSB Chair Andrew Bailey. The most notable element in this report centers on Sound Practice 10, which advocates for a human-in-command model for AI initiatives in finance.

The human-in-command model

At the heart of the FSB's framework is the critical shift from a traditional oversight approach to a revised model that emphasizes high-level human oversight. The limitations of reviewing each decision made by AI systems are clear; the speed at which these systems operate makes that virtually impossible. Rather than auditing every action taken, the FSB recommends that firms concentrate on defining the boundaries of autonomy for their AI systems and establishing necessary guardrails to govern their use.

This human-in-command directive is especially pertinent when applying AI in financial sectors where transactions involve substantial sums. Under Sound Practice 10, financial institutions must now enforce stricter controls to promote accountability and mitigate systemic risk. These requirements demand that human approval or dual authorization must be in place for all transactions surpassing certain monetary thresholds.

Moreover, the framework entails ensuring restricted access to payment systems alongside maintaining detailed audit trails for every transaction initiated by AI agents. As institutions scale their operations, the FSB anticipates that organizations will employ AI-in-the-loop monitoring—where AI systems are utilized to oversee the functions of other AI systems—compounding human oversight capabilities.

Impact of non-binding frameworks

While the guidelines outlined by the FSB do not have the power of law, their potential influence on operational practices is significant. National regulators such as the Federal Reserve, the Office of the Comptroller of the Currency (OCC), the Federal Deposit Insurance Corporation (FDIC), the Securities and Exchange Commission (SEC), and the Consumer Financial Protection Bureau (CFPB) typically absorb these international standards into their local regulatory frameworks.

In comments made by Federal Reserve Governor Michelle Bowman in July 2026, she conveyed that regulators would align expectations with existing frameworks while enhancing their focus based on materiality related to specific use cases. Legal authorities stress the importance of firms benchmarking their governance and compliance procedures against these emerging standards.

The arrival of the FSB's guidelines coincides with ongoing legislative stagnation. For instance, the Warner AI AGENT Act, designed to classify AI agents as fiduciaries, has faced delays in the Senate. Similar bills, like the Hawley-Murphy AI Agent Accountability Act, await committee discussions. Amid this legislative vacuum, the industry has taken charge, setting its operational norms.

The question of liability in agentic AI

The legal complexities surrounding AI agents continue to be pressing, particularly regarding liability. The FSB's guidance fosters internal governance; however, it does not address external legal accountability for the actions of these autonomous systems. The crux of the difficulty lies in determining who is responsible when an AI agent's actions—such as erroneous purchases, bookings, or communications—harm individuals or entities.

This question is amplified by the notion of developer liability versus user liability. Current consumer-oriented AI tools such as Meta's Muse and xAI's GrokBot operate under vendor-defined liability conditions, leaving ambiguity in accountability. Additionally, debates around the "tools-not-actors" doctrine challenge the legal standing of AI agents, prompting conversations regarding where to place responsibility.

As highlighted in our detailed Policy Radar, the legal status of agentic AI remains an urgent topic of discussion. While firms work to align with FSB’s supervisory expectations, they must also navigate the jurisdictional gray areas where existing laws struggle to encompass the innovative technology space.

Moving forward with AI governance

The evolving regulations surrounding agentic AI are signifying a crucial juncture for the financial sector. The FSB's efforts to establish a sound practical framework reflect a growing recognition among global regulators of the need for accountability and oversight in the domain of financial technology. To stay ahead, organizations must proactively work on aligning their governance structures with these new expectations.

By establishing audit trails, dual authorization protocols, and enhanced human oversight of AI systems, the financial sector can build a resilient framework that balances innovation with the essential safeguards required in a rapidly changing landscape.

In this environment of uncertainty, the emphasis must remain on the development of frameworks that can react to the pace of technology while protecting consumers and fostering sound financial practices. As the regulatory landscape evolves, the importance of structured governance frameworks will only grow, presenting firms with the challenge and opportunity to lead in this new era of AI governance.

Frequently asked questions about supervisory architecture for agentic AI

What is Sound Practice 10?
Sound Practice 10 refers to the Financial Stability Board's guideline that mandates a human-in-command approach for AI systems in the financial sector, emphasizing oversight and control over autonomous actions.

Why are the FSB's practices considered non-binding?
While the FSB's guidelines are influential, they are not legally enforceable, allowing regulators at the national level to interpret and adapt them according to their jurisdiction.

How will these practices affect future legislation?
The FSB’s framework could inform future regulatory decisions and policies, potentially shaping how lawmakers address AI governance and liability issues as the technology continues to evolve.