KESHO PARTNERS
Financial documents and calculator being reviewed

Financial services

Banking AI Readiness Guide

Most banks now use AI. Far fewer can show how it is governed. Supervisors are not waiting for new laws to close that gap, and neither should boards.

By Kesho Partners

9 minute read

A bank does not need an AI law to be held to account for its AI. Model risk, fair lending, outsourcing, data protection and conduct rules already apply from the first day a model touches a customer. The question for a board is not whether AI is regulated. It is whether the bank can show, today, what AI it runs, who owns it and how each system is checked.

AI is everywhere in banking. Governance is not.

Adoption is no longer the story. Three in four UK financial firms used AI in 2024, and more than 85% of large banks under European supervision use it in some form. In Kenya, the Central Bank's 2025 survey of 125 institutions found half had adopted AI, rising to two-thirds of commercial banks.

The gap is in control. Only 41% of Kenyan adopters had an AI policy. Most institutions placed themselves at the lowest level of maturity, and 93% asked the Central Bank for guidance. In the UK, almost half of firms reported only a partial understanding of the AI they use.

In plain terms: most banks are running AI they cannot fully describe.

Sources: [1], [2], [3]

Regulators are using the rules they already have

The enforcement record so far uses existing law, not AI law.

Fair lending.
In July 2025 the Massachusetts Attorney General settled with student lender Earnest for US$2.5 million over allegations that its AI underwriting risked unfair outcomes for Black, Hispanic and non-citizen applicants. Earnest denied the allegations. The remedy was governance: written policies, testing, an inventory and an oversight team.
Explaining a decline.
US consumer finance guidance states that a lender using complex models must still give specific and accurate reasons when it refuses credit. A generic reason that is not the real one does not meet the rule.
Credit scores as decisions.
In the SCHUFA case, Europe's highest court ruled that producing a credit score can itself be an automated decision under data protection law when a lender relies heavily on it.

The lesson is consistent. If a model influences a decision about a customer, the bank must be able to explain it, test it for fairness and show who is accountable.

Sources: [4], [5], [6]

Where supervision is heading

The direction is settled even where the text is not. Supervisors have converged on the same expectations: board accountability, an inventory, controls in proportion to materiality, validation, human oversight and control of third parties.

How major supervisors are treating bank AI, November 2026
JurisdictionApproachWhat it means
InternationalFinancial Stability Board sound practices, final October 2026A global baseline boards can reference.
European UnionAI Act; credit scoring of individuals is high-risk from December 2027Build the inventory and evidence now.
United KingdomNo new AI rules; existing conduct and model risk rules applyA named senior manager is accountable.
United StatesRevised model risk guidance, April 2026, leaves generative AI and agents out of scopeThe bank's own governance is the standard.
SingaporeDedicated AI risk management guidelines, October 2026The most complete template available.
KenyaCentral Bank folding AI into risk management guidelines, draft September 2026; data protection law already appliesExpect AI questions in ordinary supervision.

In Kenya, a separate national AI Bill was rejected by the Senate ICT committee in October 2026. That does not reduce a bank's exposure. The Central Bank's route, adding AI to prudential supervision, is the one that matters.

Sources: [7], [8], [9], [10]

Five questions every bank board should ask

  1. Show me every AI system we use, with an owner and a risk rating. If the list takes a month to produce, governance is not working.
  2. For our most material credit or fraud model, who validated it, when, and what did they find? A validation that found nothing is worth questioning.
  3. If a customer asks why they were declined, what do we say, and is it the real reason?
  4. What happens if our main model or cloud provider is unavailable for a day? Dependence on a few providers is a risk regulators now name.
  5. Which AI uses have we decided not to pursue, and why? A bank with no "no" list has no risk appetite for AI.

What readiness looks like in practice

Readiness is not a policy document. It is the ability to produce evidence on request: an inventory that reconciles with procurement, validation reports for material models, fairness testing on decisions about people, contracts that give the bank rights over vendor models, and board reporting on performance and incidents rather than a count of use cases.

It also means defending customers against AI. Fraudsters now use cloned voices and video. In 2024 a company employee in Hong Kong paid out about US$25 million after a video call in which every other participant was a deepfake.

Sources: [11]

Where to start

Start with the inventory, because every other control depends on it. Then rate each system by materiality, validate the highest-rated first, and test decisions about customers for fairness and explainability. Take the results to the board as decisions with owners and dates.

Our guide sets this out in full: ten areas of assessment, the evidence to request, the red flags that matter, 42 questions for management and a 100-day plan.

The Bank Executive's Guide to AI Readiness and Governance

Sources

  1. Bank of England and FCA. Artificial intelligence in UK financial services, 2024.
  2. ECB Banking Supervision. Speech, February 2026.
  3. Central Bank of Kenya. Survey on Artificial Intelligence in the Banking Sector, 2025.
  4. Massachusetts Attorney General. Settlement with student loan lender over AI use, July 2025.
  5. CFPB. Circular 2023-03.
  6. A&O Shearman. CJEU rules that a credit score constitutes automated decision-making.
  7. Financial Stability Board. Sound practices for responsible adoption of AI, October 2026.
  8. Federal Reserve. SR 26-2, April 2026.
  9. Monetary Authority of Singapore. Guidelines on AI risk management, October 2026.
  10. Techweez. Senate committee rejects AI Bill, October 2026.
  11. South China Morning Post. Arup confirmed as victim of HK$200 million deepfake scam, May 2024.

This article is general information, not legal advice.

Related service

Financial services

Independent AI readiness and governance assessments for banks and lenders, from inventory to board report.

Explore the service