What does FSB 2017 AI/ML report say about AI in banking?

What are the key points of FSB 2017 AI/ML report?

• Published 1 November 2017, before generative AI; covers 'traditional' ML such as credit scoring, trading signals and anti-fraud.
• Use cases examined: credit quality assessment, insurance pricing and marketing, automated client interaction, capital optimisation, trading signal generation and execution, compliance and fraud detection.
• Warns that shared data sources and similar models across firms can create unexpected interconnectedness and correlated behaviour.
• Identifies reliance on a small number of third-party AI/data providers as a potential source of new systemically important players outside the regulatory perimeter.
• Highlights model opacity: lack of interpretability could make it hard for firms and supervisors to understand how AI affects markets at a macro level.
• Calls for adequate testing on unbiased data, feedback mechanisms, and continued monitoring as adoption expands.

What did FSB 2017 AI/ML report change for banks?

It established the vocabulary supervisors still use for AI risk — third-party concentration, model correlation, opacity — nearly a decade before the 2024–2026 workstream, and is the reference point for the FSB's claim that many AI vulnerabilities are not new.