Artificial Intelligence

GOVERNING AI IN KENYA’S BANKING SECTOR: INSIDE CBK’S DRAFT GUIDANCE NOTES ON ARTIFICIAL INTELLIGENCE.

Euro coin on stock chart. Financial investment concept. Close up. · Image used is from envato.com

Introduction.

The Kenya national Artificial Intelligence(AI) strategy recognises AI as a driver of change in diverse sectors, including in finance.1 Artificial intelligence is increasingly becoming embedded within financial institutions and their operations. Algorithms are assisting in various functions, including identifying fraudulent transactions, assessing creditworthiness and managing operational risk.2The Central Bank of Kenya(CBK) survey, conducted in March 2025, confirmed this development. The survey aimed to assess the current landscape of AI adoption within financial institutions, its benefits, emerging risks, and future trends. The survey collected data from commercial banks, mortgage finance institutions, credit reference bureaus, microfinance banks, and digital credit providers. The focus areas were AI and data strategy, institutional AI activities, AI risk management and the impact and challenges of AI.3

The survey revealed that only 50% of respondents had adopted artificial intelligence, with AI maturity levels uneven across institutions. Some institutions had developed governance structures, AI implementation structures, and AI implementation strategies, while others were still integrating AI within broader digital transformation programmes. The survey identified gaps in governance arrangements accompanying AI adoption; among institutions that had adopted AI, 59% had not implemented AI policies. It also highlighted the need for guidance on governance and compliance, risk management frameworks and incident management reporting. Overall, 93% of the respondents recommended that CBK issue guidance on AI.4

CBK’s draft Guidance Note on Artificial Intelligence in the Banking Sector seeks to translate AI governance principles into practical requirements for regulated financial institutions. Published for public comment on 10 September 2026, it formed part of a broader review of CBK’s supervisory framework, with comments invited by 7 November 2026.5 This blog post therefore examines the published draft; the requirements discussed remain proposals subject to finalisation. Notably, the draft shows how sector-specific supervision could link AI adoption to institutional accountability and consumer protection.

Overview of the Draft Guidance Notes.

The draft Guidance Note would apply to all institutions licensed under the Banking Act and the Microfinance Act, specified credit reference bureaus, money remittance providers and non-deposit-taking credit providers licensed under the Central Bank of Kenya Act, and non-operating holding companies where the covered institutions would need to seek CBK approval before deploying AI systems.6 The institutions referenced in the guidance notes will therefore need to get approval from CBK before deploying AI systems.

The draft defines its coverage by reference to specified institutional categories which can be understood as an institution-based approach, in contrast to an activity-based approach that applies requirements to specified activities across providers.7

The guidance notes treat AI as an enterprise wide risk management issue rather than a technology concern.8The implication is that CBK is signalling that AI is not just another onboarded technology; instead, it can affect every part of a financial institution’s risk profile. Some of the risk profiles include credit and cyber risk, operational risk and reputational risk.9 Consequently, the guidance notes integrate AI oversight into existing governance, compliance, operational, risk, cybersecurity, and reputational risk structures. This acknowledges that AI systems can affect multiple areas simultaneously; they may generate legal, ethical, operational, and consumer-protection risks.10 The guidance promotes ethical and responsible AI use, protects consumers from unfair or opaque AI decisions, improves governance and accountability, and addresses risks such as algorithmic bias, privacy breaches, cybersecurity threats, hallucinations, explainability challenges and autonomous decision-making risks. It seeks to support innovation while maintaining prudential oversight.11

The draft guidance notes assign ultimate responsibility for AI governance to the board, with assistance from senior management.12 Financial institutions will be accountable for all decisions made with the assistance of AI systems, even when those systems are developed or supplied by third-party vendors.13 This means that directors are expected to understand and supervise AI-related risks. It places oversight squarely within the board, moving away from the narrative that AI governance is solely the responsibility of IT departments. To reinforce accountability, institutions are expected to establish governance structures including AI governance, data governance and independent model risk management functions.14

It is mandated that the board have a data governance committee, an AI governance committee and a model risk committee.15 The data governance committee will be responsible for implementing the company’s data governance strategy, policy and framework, while the AI governance committee is expected to implement AI initiatives. These two committees may work as one depending on the size of the company. 16The model risk committee will be primarily tasked with identifying, assessing, controlling, managing, reporting and remediating model risks through the AI lifecycle. The committee may coordinate with the data and AI governance committees on matters relating to data used in machine learning models and AI systems. The AI and data governance committees shall not perform independent validation or approval functions assigned to the model risk committee.17 In essence, these committees will be responsible for implementation, monitoring and reporting to the board on their various functions.

Data is identified as the foundation of effective AI systems. Through data, systems are trained to identify and predict patterns.18 A system is as good as the data it trains on. Consequently, institutions are expected to implement robust controls around data quality, security, lineage, privacy and retention.19Data governance is therefore embedded at the centre of effective AI governance.20 It recognises that flaws in underlying datasets often translate into flawed AI outcomes, including inaccurate predictions, biased decisions and discriminatory treatment of customers.21 To ensure effective data governance, the notes require companies to have effective data governance committees to oversee the implementation of data governance strategies, policies, and frameworks. This committee will make periodic reports to the board and senior management.22

In alignment with its customer-centricity principle, the guidance note includes express provisions on consumer protection. To ensure they uphold their customer protection obligations, particularly fairness and non-discrimination, financial institutions should ensure that their AI outputs do not result in unfair, unreasonable or discriminatory treatment.23 They are also obligated to have mechanisms in place that accommodate consumers’ right to seek redress, explanations, and to challenge AI-assisted decisions where they have reason to believe they were not treated fairly. As such, financial institutions would need to be proactive in ensuring AI outcomes are contextual, clear and accurate, thereby ensuring that consumers are not left at the mercy of automated systems.24

In its guidance notes, the CBK does not limit itself to the present systems in use as guidance notes anticipate that generative AI will also find root in the industry. It takes note of risks such as hallucinations, intellectual property concerns, confidentiality breaches and employee awareness and training.25 It therefore requires that, in the event that an institution decides to introduce generative AI into its pre-existing system, it should also update its governance framework to address the accompanying issues as listed above.26 Agentic AI systems are to be classified as high-risk systems that need enhanced controls and oversight. Human oversight has been made mandatory in the operation of Agentic AI systems.27

Observable Gaps in the Draft Guidance Note

The guidance note takes an institutional approach. It mainly focuses on the institutions it regulates. This is as opposed to taking an operational approach, where it would regulate entities that partake in financial activities that are regulated by CBK. For instance, the most significant omission is the exclusion of payment service providers (PSPs). Solely subjecting certain institutions to the guidelines while excluding others is likely to result in regulatory fragmentation and inconsistent consumer protection standards across the financial sector.

As stated earlier, the draft guidance notes recognise data as the bedrock of AI, ensuring that data processing requires a lawful basis under the Data Protection Act. CBK calls upon financial institutions to have clear bases when processing data.28 It places an obligation on institutions to get consent before processing.29 There is a lack of consistency with the existing data protection laws since, depending on the circumstances, one basis may be more befitting.30 The data protection Act does indeed provide for consent as the primary basis, but the general regulations provide room to process data without consent in the event that the processing is necessary for other reasons, including legitimate interest and the performance of an obligation.31

Several provisions in the guidance notes need clarification; they have been barely explored. One example is the question of relevant AI incidents. Notification obligations have been placed on institutions where relevant AI incidents occur. The guidance notes define relevant AI incidents as incidents that significantly affect critical systems, disrupt services, or compromise data integrity, requiring notification to the Central Bank of Kenya.32 However, it fails to explain the metric for measuring significance; it is unclear at what level an incident affecting critical systems, disrupting services, or compromising data integrity is deemed significant enough to warrant notification.

While it notes AI as a source of risk, it does not extensively address AI as a tool for malicious actors. Increasingly relevant threats include AI-generated fraud schemes, deepfake-enabled identity fraud, synthetic identity attacks, and AI-assisted social engineering that may warrant more detailed guidance. Express provisions on how to navigate these threats and risks would have been key to guiding vulnerable financial institutions.

Conclusion

Despite these shortcomings, the draft Guidance Note represents a significant milestone in Kenya’s emerging AI regulatory landscape. Its strengths lie in its comprehensive governance architecture, strong consumer protection requirements, lifecycle-based risk management approach and recognition of emerging technologies such as generative and agentic AI. The framework sends a clear message that innovation and accountability must evolve together.

The more difficult challenge will be ensuring that regulation keeps pace with the realities of Kenya’s broader digital financial ecosystem. As AI adoption extends beyond traditional banks to fintechs, payment service providers and other innovative market participants, the draft Guidance Note positions Kenya among a growing group of jurisdictions seeking to shape the future of AI in financial services before risks become crises, rather than after.

1Ministry of Information, Communications and the Digital Economy, Kenya AI Strategy 2025–2030 (2025) <https://ict.go.ke/sites/default/files/2025-03/Kenya%20AI%20Strategy%202025%20-%202030.pdf> accessed 13 October 2026.

4 Central Bank of Kenya, Survey on Artificial Intelligence in the Banking Sector (July 2025) ss 2.2, 3.1.1 and 3.1.4.

5 Central Bank of Kenya, ‘Invitation for Comments from the Public on the Draft Revised Prudential Guidelines, Risk Management Guidelines, Guidance Notes and the Domestic Systemically Important Banks Framework’(Public Notice, 10 September 2026) <https://www.centralbank.go.ke/2026/09/10/draft-cbk-prudential-guidelines-risk-management-guidelines-guidance-notes-and-framework-for-domestic-systemically-important-banks/> accessed 10 November 2026.

7 Claudio Borio, Stijn Claessens and Nikola Tarashev, ‘Entity-based vs activity-based regulation: a framework and applications to traditional financial firms and big techs’ (FSI Occasional Papers No 19, 3 August 2022) <https://www.bis.org/publications/fsi-paper-19-entity-based-vs-activity-based-regulation-framework-and-applications-traditional-financial-firms-and-big-techs> accessed4 November 2026

Comments

Leave a comment

Your email address will not be published.

Your message may be processed by a trusted service to keep this site safe.