
The FSB's Final AI Report Puts One Decision Ahead of the Rest
The Financial Stability Board published its final report, Sound Practices for Responsible Adoption of Artificial Intelligence (AI), on 15 October 2026, setting out 12 non-binding practices financial institutions are encouraged to consider as they adopt AI. Compliance Week reported the release on November 4, 2026, describing 12 "sound practices" for banks and other major financial services firms. The FSB's final report page organises them in two groups: organisation-wide AI governance, and risk management throughout the AI lifecycle.
But here is the part a list of twelve headings hides. Five of the twelve practices (3, 5, 6, 9 and 10) expressly turn on the materiality and risk of an individual AI use case. Model selection, performance testing, and human oversight each turn on materiality and risk, which Practice 5 says firms assess at the inception stage and thereafter.
A firm that has not classified its AI use cases has no documented basis for deciding how much testing or oversight each one receives.
Non-Binding, but Addressed to the Board
The practices are not law. The FSB's June 2026 consultation press release states they are "not intended to establish an international standard" or "to impose a prescriptive approach," and Regulation Tomorrow's summary of the final report describes them as non-prescriptive and not a binding international standard. The same summary places practices 1 to 4 under governance and the subsequent eight under the AI lifecycle, "from inception to retirement." The practices function as a reference framework, not a compliance deadline.
What the practices do carry is an audience. The final report page says the board and senior management of financial institutions "are strongly encouraged to reference the sound practices" as they consider business strategy, technology adoption, and risk management. The June consultation quoted Michelle Bowman, Chair of the FSB Standing Committee on Supervisory and Regulatory Cooperation and Vice Chair for Supervision of the Federal Reserve Board, who said the final report would be issued "as a US G20 deliverable." The final report page lists it under Reports to the G20.
Comments closed on 22 July 2026, and the FSB published the public responses on 6 August 2026. Respondents included JPMorgan Chase, Visa, Mastercard, Wise, Binance, the Managed Funds Association, and ICI Global.
Inventory Is Not Classification: The Materiality Call Behind Five Practices
An AI inventory and an AI materiality assessment are different things. An inventory tells you which tools are in use. A materiality assessment tells you which of them can hurt clients, the balance sheet, or the firm's regulatory standing, and by how much. Practice 3 asks for both: risk management frameworks that "include processes for documentation of AI use, as well as for materiality and risk assessment."
Practice 5 sets the timing: an "effective and systematic approach to assess the materiality and risk of AI use cases at the inception stage and thereafter." Three later practices also turn on materiality and risk:
- Practice 6 (Selection process) weighs "the materiality and risks of AI use cases" when choosing models or systems.
- Practice 9 (Performance management) evaluates performance "proportionately to their materiality and risk."
- Practice 10 (Human oversight) ties oversight to "the materiality, risk, autonomy, complexity, and explainability" of each use case.
The practical consequence is proportionality with a paper trail. A tool that summarizes internal meeting notes and a model that informs client portfolio decisions sit at different points on that scale, and Practices 9 and 10 tie performance testing and human oversight to materiality and risk. Regulation Tomorrow notes the final report includes case studies showing how some practices have been applied proportionately.
The FSB's final report page says rapid advances, "particularly in generative AI (GenAI) and agentic AI," generated the interest behind the report. Agentic tools act with more autonomy, and autonomy is one of the factors Practice 10 ties human oversight to, so Practice 10 points to more human oversight for agentic tools than for a comparable tool that does not act on its own.
Practice 12 Describes Your Vendors, or Describes You
Practice 12 asks financial institutions to manage "risks from third-party AI dependencies with a focus on performance, transparency, data quality, supply chain and concentration risks, and business continuity." For an adviser or fund manager running on a vendor's model, that is a list of questions to put to the vendor. For a fintech selling AI into banks or asset managers, it is a description of the diligence the customer's board is encouraged to run on you.
Practices 7 and 8 sharpen the same point. Practice 7 calls for data that is "fit-for-purpose for training, testing, and using AI (i.e. accurate, complete, consistent, reliable, secure)." Practice 8 asks firms to understand differences in explainability and, "if appropriate and feasible," adopt more explainable AI or consider compensating controls.
A vendor that cannot describe its training data or explain its outputs may push those compensating controls back onto the buying firm. That is a cost, and it belongs in the procurement decision, not after it.
What Your Leadership Team Should Decide This Quarter
The practices are voluntary. Here is how to sequence them.
Governance (Practices 1 to 4)
- Name an owner. Practice 2 calls for clear roles and responsibilities; decide which executive owns the AI framework and reports on it to the board.
- Tie AI to risk appetite. Practice 1 asks the board and senior management to align AI adoption with the firm's business model, risk appetite, and strategy.
- Schedule the revisit. Practice 4 expects oversight and capabilities to adjust as AI evolves, so set a review cadence now.
Lifecycle (Practices 5 to 12)
- Classify every use case. Build a written materiality and risk methodology and apply it at inception, as Practice 5 describes.
- Map testing and oversight to the classification. Higher-materiality use cases get more performance testing under Practice 9, and Practice 10 scales human oversight to materiality, risk, autonomy, complexity, and explainability.
- Re-paper vendor diligence. Add Practice 12's five focus areas to AI vendor questionnaires and contract terms.
- Fold AI into cyber exercises. Practice 11 includes incorporating AI cyber and ICT risk scenarios into tests and exercises where appropriate.
Key Takeaways
- The FSB's 12 AI sound practices are final but non-binding. The final report was published on 15 October 2026 after a 10 June 2026 consultation, and the FSB strongly encourages boards and senior management to reference it.
- Materiality is the gating decision. Practices 3, 5, 6, 9 and 10 each turn on the materiality and risk of an individual AI use case.
- An AI inventory is not enough. Practice 3 pairs documentation of AI use with a materiality and risk assessment process.
- Vendors are in scope through their customers. Practice 12 directs financial institutions to manage third-party AI risks covering performance, transparency, data quality, supply chain and concentration risks, and business continuity.
Advisers, fund managers, and fintechs that deploy AI or sell it to financial institutions may want their AI use-case inventory, materiality methodology, and vendor contracts reviewed against Practices 3, 5 and 12. FinTech Law's RegTech counsel does that work, and you can start through our contact page.
This blog post is for informational purposes only and does not constitute legal advice. No attorney-client relationship is formed by reading this content. If you need legal advice, please contact a qualified attorney.
FinTech Law fixed engagements: FinTech Startup Regulatory Path ($1,500) · Emerging Manager Launch Kit ($7,500) · Regulatory Path Outline ($7,500) · RIA/ERA Registration ($8,500) · Private Fund Formation (from ~$18,000) See the catalog.
Fractional General Counsel · from $7,500/qtr