Tool · Governance checklist

Enterprise AI Governance Checklist

Forty controls in eight domains. Answer Yes, Partial or No; the score and maturity band update as you go, and the page prints as a document you can table at the next risk committee. Every control cites its source.

Scoring. Yes = 2, Partial = 1, No = 0. Each domain scores out of 10; the total is out of 80 and shown as a percentage. Bands: Foundational below 40%, Developing 40–69%, Established 70–89%, Assured 90% and above. The bands are our editorial scale, not a regulatory rating. Answers are kept only in your browser.

Without JavaScriptThe checklist still works on paper: print this page, mark each control, add 2 for Yes and 1 for Partial, and compare the total out of 80 with the bands above.

Overall

Foundational

0 of 80 points · 0 of 40 answered

By domain

Domain	Score	Meter
01 Accountability and policy	0/10
02 Model inventory and lifecycle	0/10
03 Data protection (CDPA and SI 155)	0/10
04 Security and access control	0/10
05 Vendors and third parties	0/10
06 Human oversight and automated decisions	0/10
07 Monitoring, audit trail and incidents	0/10
08 Skills, culture and change	0/10

01 · Accountability and policy

0 / 10

01 A board-approved AI policy exists and was reviewed within the last 12 months. Act s.24; RBZ PS 01-2024 para 7.2.8

Yes Partial No

02 A named executive is accountable for AI, with the responsibility in their objectives. RBZ PS 01-2024 para 8.4

Yes Partial No

03 An AI risk appetite statement says which decisions may be automated and which may not. RBZ PS 01-2024 section 2.9

Yes Partial No

04 Every AI use case is classified by impact (customer-affecting, employee-affecting, operational, internal) before build. NIST AI RMF 1.0, Map function

Yes Partial No

05 The board or its risk committee receives an AI and model report at least quarterly: register changes, incidents, validation findings. RBZ PS 02-2023 para 2.1.5

Yes Partial No

02 · Model inventory and lifecycle

0 / 10

06 A complete model register exists, including vendor-embedded and hosted models, with owner, purpose, data, hosting location and materiality. RBZ PS 02-2023 Definitions (model register/inventory)

Yes Partial No

07 Material models are identified and receive board-level oversight. RBZ PS 02-2023 paras 2.1.3–2.1.4

Yes Partial No

08 Material models are independently validated before production and re-validated on change. RBZ PS 02-2023 paras 2.4.5, 2.4.17

Yes Partial No

09 Models, prompts and retrieval indexes are version-pinned under change control with a model log. RBZ PS 02-2023 Definitions (model log)

Yes Partial No

10 A retirement procedure covers model decommissioning and deletion of associated data and indexes. Act s.13 (retention no longer than necessary)

Yes Partial No

03 · Data protection (CDPA and SI 155)

0 / 10

11 The organisation holds a data controller licence in the correct tier and renews at least three months before expiry. SI 155 s.4–6

Yes Partial No

12 A Data Protection Officer is appointed, certified and notified on Form DP2, with continuing professional development funded. SI 155 s.10(1), s.12–13

Yes Partial No

13 Processing activities, including AI systems, are notified to the Data Protection Authority. SI 155 s.10(2)(a); Act s.20

Yes Partial No

14 Personal data is classified at ingestion; sensitive, genetic, biometric and health data are processed only with written consent; biometric processing is notified. Act s.11–12; SI 155 s.10(2)(d)

Yes Partial No

15 Cross-border transfers are assessed for adequacy, a legal basis is documented, the Authority is notified, and processors are under written contract. Act s.18(8), s.28–29; SI 155 s.10(2)(c), s.10(4)(f)

Yes Partial No

04 · Security and access control

0 / 10

16 All model calls pass through a gateway the organisation operates; no application holds a vendor API key. Act s.18(4); RBZ Guideline para 6.4

Yes Partial No

17 Retrieval enforces the requesting identity's entitlements; the model never decides access. Act s.18; RBZ Guideline access-management requirements

Yes Partial No

18 Prompt-injection and output-handling controls are in place: no shell or raw-SQL tools, parameterised tools, screened retrieved content. OWASP Top 10 for LLM Applications 2025 (LLM01, LLM05)

Yes Partial No

19 Secrets, logs and indexes are encrypted in transit and at rest; administrative access is nominative with multi-factor authentication. RBZ Guideline para 6.4 (core security controls); SI 155 s.16

Yes Partial No

20 The AI surface is included in the annual penetration test and findings are tracked to closure. RBZ Guideline section 4 (testing), para 6.4

Yes Partial No

05 · Vendors and third parties

0 / 10

21 A due-diligence file exists for every AI vendor: security policies, certifications, sub-processors, data handling. RBZ Guideline para 5.3; Act s.18(7)

Yes Partial No

22 Contracts include audit rights, data segregation, portability, exit arrangements and incident cooperation. RBZ Guideline para 5.6; Act s.18(8)

Yes Partial No

23 Vendor licensing and international sanctions screening is completed and recorded. RBZ Guideline para 6.4(b)

Yes Partial No

24 A current third-party inventory exists and substitutability of critical AI providers is assessed. RBZ Guideline paras 5.7, 5.14

Yes Partial No

25 For banking institutions: the Reserve Bank was informed early of cloud outsourcing of critical functions and prior written approval was obtained before go-live. RBZ Guideline paras 5.12, 6.4(a)

Yes Partial No

06 · Human oversight and automated decisions

0 / 10

26 An inventory of decisions with legal or similarly significant effects exists, and none is taken solely by automated processing without consent or a legal basis. Act s.25; SI 155 s.10(3)

Yes Partial No

27 A human route to contest an automated outcome exists and is communicated to data subjects. Act s.14, s.25

Yes Partial No

28 Agent tools are classified read, propose or write; write actions require a named human approval enforced in the tool gateway. RBZ Guideline para 6.4(c) (evidence of approvals and audit trails)

Yes Partial No

29 Children's data is never subject to automated decision-making that affects their rights. SI 155 s.10(5)(f)

Yes Partial No

30 Customer-facing notices disclose AI use and purposes at the point of collection. Act s.15

Yes Partial No

07 · Monitoring, audit trail and incidents

0 / 10

31 Prompts, retrieved record identifiers, tool calls and outputs are logged with the initiating identity and retained under access control. Act s.14; RBZ PS 01-2024 para 7.2.8

Yes Partial No

32 Drift, quality and bias are monitored with thresholds and named owners. RBZ PS 01-2024 paras 7.2.7–7.2.8

Yes Partial No

33 The breach path has been tested by drill: 24 hours to the Authority on Form DP3, 72 hours to affected people where high-risk, and 3 hours to the Reserve Bank for regulated institutions. Act s.19; SI 155 s.17(1)–(3); RBZ Guideline para 4.30

Yes Partial No

34 A breach register is maintained and the organisation can answer information requests within 14 days and report within 21 days. SI 155 s.17(4)–(5)

Yes Partial No

35 AI incidents feed the operational risk event database and board reporting. RBZ PS 01-2024 section 7.3

Yes Partial No

08 · Skills, culture and change

0 / 10

36 The board and executive received an AI briefing within the last 12 months. RBZ Guideline para 7.1

Yes Partial No

37 An acceptable-use standard tells staff which AI tools are approved and which data may never be entered. SI 155 s.16(2)(b)

Yes Partial No

38 Role-based training is in place for model owners, validators, the DPO and the CISO. RBZ Guideline section 7; SI 155 s.13(2)

Yes Partial No

39 The whistleblowing or concern route explicitly covers AI misuse. Act s.31

Yes Partial No

40 An annual governance review is scheduled and changes in law and regulator guidance are tracked. Act s.18(6) (Authority may issue standards)

Yes Partial No

Discuss the result in a briefing

MethodControls were derived from the eight domains on the governance page. Citations refer to the Data Protection Act, Act 5 of 2021 (cited in regulations as the Cyber and Data Protection Act [Chapter 12:07]), Statutory Instrument 155 of 2024, RBZ Prudential Standards No. 01-2024/BSD and No. 02-2023/BSD, the RBZ Cybersecurity and Resilience Guideline (August 2025), NIST AI RMF 1.0 and the OWASP Top 10 for LLM Applications 2025. Non-bank organisations should read RBZ items as good practice. Last reviewed September 2026.