FSB Sets 12 Principles for AI in Finance as Banks Face Tougher Accountability

The Financial Stability Board (FSB), the international body that coordinates financial regulation among major economies, finalized 12 principles on October 15, 2026 to guide how financial firms adopt artificial intelligence. The principles cover everything from board-level oversight to performance validation, human supervision, cybersecurity and reliance on outside technology vendors.

What the 12 Principles Ask of Banks

The FSB released a draft in June and received about 160 comments before finalizing the report. Boards and senior management must make sure AI strategy fits the firm's business model and risk appetite, and decisions and responsibilities must be documented. Banks are asked to assess purpose and risk before deployment, keep measuring performance in live operation, manage data quality and security, and be able to explain AI judgments. Human oversight should match each system's risk, autonomy and complexity rather than treating every output the same way. Dependence on external AI providers is its own risk item, covering transparency, data quality, supply chain stability and business continuity. The principles are international guidance, not binding rules.

Experience from European banks shows why this matters. Agentic AI, which can use tools and carry out tasks rather than only draft text, raised productivity in customer due diligence and source-of-funds checks. But approval of a first use case took 6 to 8 months at some institutions, and one bank's repeated tests showed answer consistency of 72% to 100%. Security threats such as prompt injection, where malicious instructions are hidden in documents an AI reads, also demand tight access controls.

Regulators are moving too. The UK's Financial Conduct Authority is testing AI with eight firms, including Barclays, Experian and UBS, and plans to publish an evaluation in the first quarter of 2027. Under the EU AI Act, high-risk obligations under Annex III apply from December 2, 2027, and a credit-scoring AI is classified differently from a fraud-detection tool.

Why It Matters

As AI moves from drafting reports to executing banking tasks, supervisors are shifting focus from model accuracy to autonomy, data access, audit trails and who is accountable. Banks that build risk-tiered approval and continuous monitoring early will likely move faster than those approving each project from scratch.

Read the original article in Korean: K trendy NEWS

댓글

이 블로그의 인기 게시물

Incheon Explores Cleaner Tap Water and Stream Restoration

Limhana's 'Replenish': Silver Matter, Black Cloth and the Return of Balance