Banks told to fix AI controls
Regulators are forcing banks to tighten data, security and decision‑rights as ‘agentic’ AI tools move into core workflows — banks must upgrade audit trails and model governance now, not later reported and analysts warn operational fixes are urgent noted. For quant engineers and data scientists that means more documentation, validation and explainability work baked into pipelines.
Pymnts (pymnts.com) reported on March 16, 2026 that banks are scrambling to control “non‑human identities” — machine credentials, tokens and service accounts — as agentic AI moves from pilots into production. The Financial Conduct Authority (FCA) (fca.org.uk) published an AI Update on April 22, 2024 (last updated Dec 3, 2025) saying it will apply “close scrutiny” to firms’ systems and processes, signalling more intensive supervisory reviews of AI tooling. The Bank of England and the Prudential Regulation Authority (PRA) (bankofengland.co.uk) set out their strategic approach on April 22, 2024, emphasising board oversight and risk frameworks as prerequisites for scaling AI in regulated firms. The Bank for International Settlements’ Consultative Group on Risk Management issued a governance report in January 2025 that recommends embedding AI into a three‑lines‑of‑defence model and explicit risk registers for model drift and hallucinations. (bis.org) The Basel Committee agreed to analyse ICT risk management and digitalisation in its 2025–26 work programme after meetings on March 12–13, 2025, raising the prospect of cross‑jurisdictional supervisory expectations for banks’ AI controls. (bis.org) KPMG’s guidance lists concrete validation requirements — fairness testing, explainability checks, robustness and challenger‑model workflows — as part of model validation and audit readiness for AI in finance. (kpmg.com) Commercial vendors already sell model‑governance platforms that centralise model inventories, versioning and examination‑ready audit trails, reflecting how banks are operationalising regulator expectations today. (ituring.ai) Industry research shows adoption is accelerating — a Google Cloud‑cited survey found 53% of financial institutions use AI agents in production and 49% plan to allocate at least half of future AI budgets to agentic tools — reinforcing why operational fixes for data, access controls and governance are now a compliance priority. (prosightfa.org)
