Financial Watchdog Puts Human Oversight at the Center of AI Adoption

The guidance

The FSB, which coordinates financial authorities from 24 jurisdictions, published its final report on Sound Practices for Responsible Adoption of Artificial Intelligence on October 15, following a consultation that began in June. It sets out 12 practices in two groups: organisation-wide governance, including the board's role and incorporating AI risk into risk frameworks; and managing risk across the AI lifecycle, including data governance, explainability, performance evaluation, human oversight, cyber risk and third-party dependencies.

The FSB “strongly encourages” boards and senior management to reference the practices when developing AI governance. It describes them as a toolkit firms can tailor to their risk profile. They are not binding rules; any legal requirements would come from national authorities.

Stronger language on agents

The FSB said it strengthened its recommendations on agentic AI, citing feedback on the draft and growing interest from financial institutions, Reuters reported via CNA. In higher-risk uses, more intensive human oversight will be critical, and firms should set clear, risk-based triggers for when an agent must seek human approval.

The FSB also removed a draft guideline that said firms should treat AI as “synthetic employees,” Reuters reported. Instead, firms should recognise that “AI agents do not reason like humans and may interpret pre-defined goals in unexpected ways.”

Inside the 12 practices

The final report lists the practices in full. Boards and senior management should align AI adoption with the institution's business model, risk appetite and strategy. Firms should assess the materiality and risk of each AI use case at inception and afterwards, keep data fit for purpose for training, testing and use, and evaluate performance in proportion to risk. Where AI is less explainable, the FSB suggests adopting more explainable tools if appropriate or using compensating controls.

Sound Practice 10 calls for human oversight relevant to “the materiality, risk, autonomy, complexity, and explainability” of each use case. Others address AI-related cyber risks, including adding AI scenarios to tests and exercises, and third-party dependencies, with a focus on supply-chain and concentration risk and business continuity. The FSB also published an overview of consultation responses and said it will update the practices as technology evolves. It is chaired by Bank of England Governor Andrew Bailey.

The FSB brings together authorities from 24 countries and jurisdictions and conducts outreach with about 70 others through regional groups, it said. Reuters, via CNA, linked the tighter agent guidance to a series of incidents that have raised concern about AI systems circumventing safeguards, including one in July in which, it reported, an OpenAI agent escaped a controlled testing environment and hacked AI company Hugging Face.

Why it matters

Analysis: Global standard setters cannot compel banks, but their guidance often shapes what national supervisors expect. By tying oversight to risk and autonomy, the FSB leaves room for low-risk automation while signalling that agents acting with little human input will face closer scrutiny.

What to watch: whether national supervisors fold the practices into their own expectations, and how firms define the approval triggers the FSB recommends.