AI practices strengthen oversight of autonomous and third-party systems
Around 160 responses informed the Financial Stability Board (FSB) revisions to its AI adoption practices, published October 15, 2026. The consultation ran from June 10 to July 22, 2026.
Accountability follows autonomous actions
The final report retains principles-based, technology-neutral and proportionate practices for financial institutions.
Sound Practice 3 strengthens documentation for accountability, traceability and reconstruction of agentic AI actions.
Practice 8 reinforces transparency for higher-risk and higher-materiality uses, including supervisory engagement.
Practice 9 adds independent validation, expert assessment of generative AI and real-time monitoring of autonomous systems.
Practice 10 expands human oversight considerations.
Practice 12 covers embedded AI and sub-outsourcing, provider concentration, granular assurances, transparency and continuity and exit strategies.
It encourages shared contractual terms, assurance tools and joint scenario testing.
Existing governance gains AI controls
Respondents supported flexibility but raised vendor transparency gaps, provider dependencies and consumer harm.
Governance Practices 1-4 clarify board and senior management roles, smaller institutions' options, cross-functional coordination and retention of human expertise.
Lifecycle Practices 5-12 retain risk-based selection, data governance and explainability.
The report adds nonbank and agentic AI case studies and expands its glossary.
It also elaborates workforce benefits, deskilling risks and documentation gaps; case studies remain illustrations, not benchmarks.
Institutional controls stop short
The report strengthens institution-level controls but excludes broader systemic risks and macroprudential concerns.
The FSB will monitor AI-related systemic risks through regular vulnerability assessments.
Added agentic AI considerations were not fully consulted on and remain emerging practices.
IN: