Amendment Notice ART-AIMS-G002 v1.1 · Supersedes v1.0 · Standards-aligned reconstruction · 4 July 2026
Public · ART-AIMS-G002 v1.1 · ISO/IEC 42001:2023 Cl. 5.3 · Gap Article 2 of 9 · unuslondon.com · 3 July 2026
G2 · Critical ISO 42001 · Clause 5.3 — Organisational Roles, Responsibilities & Authorities

The COLP Problem: Who is actually responsible for AI in your firm?

The SRA's December 2025 thematic review found that only one COLP across its sample could outline all of their responsibilities. With AI governance now part of that obligation, the accountability gap is no longer theoretical.

DocART-AIMS-G002 v1.1 StandardISO/IEC 42001:2023 Cl. 5.3 SeverityCritical Fix DocRACI-AIMS-001 Read12 min
1 COLP in the SRA sample who could outline all their responsibilities (Dec 2025)
0 UK law firms with a documented AI governance RACI as of Q1 2026
18 Fabricated AI citations in Ayinde v Haringey [2025] EWHC 1383 — no named responsible individual
5 Questions in Pemberton Capital's AI governance questionnaire — Q1 asks: "Who owns AI risk?"

When the regulator asks who owns AI risk, "we all do" is not an answer

In December 2025, the SRA published its thematic review of compliance officers — Compliance Officers: A Thematic Review. The headline finding was stark: across a representative sample of regulated firms, only one COLP could outline all of their responsibilities. The review predated the widespread deployment of generative AI tools inside those very firms.

The implication is significant for every managing partner. If the individual designated by statute to hold regulatory accountability cannot articulate their own responsibilities, the question that follows is direct: when AI-generated work product causes harm — a fabricated case citation, a confidential data disclosure to a cloud model, a client misled by automated advice — who is accountable? The answer, under both SRA obligations and ISO/IEC 42001:2023, is the same: a named role shall be accountable. The question is whether the firm has decided which role, in a controlled document, before the incident occurs.

⚠ Warning — Regulatory Enforcement Consequence

Ayinde v Haringey / Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin) — presided over by Dame Victoria Sharp P — recorded 18 fabricated AI-generated citations submitted in court filings. The judgment did not attribute the failure to "the firm." It required identification of the individual responsible for supervising the output. A firm without a documented AI governance structure — and without a named, accountable role — shall be unable to provide that identification.

A firm that cannot demonstrate assigned AI governance accountability following a judicial or regulatory inquiry is exposed to findings of systemic compliance failure under SRA Code of Conduct for Firms 2019, Rule 8.1(a).

ISO/IEC 42001:2023 Clause 5.3 provides the framework for that answer. This article explains what the clause shall require, why most law firms fail it on first assessment, and how to build an AI governance RACI that satisfies both the standard and the SRA's regulatory expectations.


What ISO/IEC 42001:2023 Clause 5.3 requires

ISO/IEC 42001:2023 — Clause 5.3 · Organisational Roles, Responsibilities and Authorities

Top management shall ensure that the responsibilities and authorities for relevant roles are assigned and communicated within the organisation.

Top management shall assign the responsibility and authority for ensuring that the AI management system conforms to the requirements of this document, and for reporting on the performance of the AI management system to top management.

Per ISO/IEC Directives Part 2, shall denotes a requirement without discretion; should denotes a recommendation; may denotes a permission. Clause 5.3 sits within Section 5 (Leadership) — it is a governance obligation, not a process recommendation. The standard requires that before any AI system is deployed, a named role shall be formally assigned responsibility for that system's conformance and performance.

Three elements shall be present in combination:

  1. Assignment — responsibilities shall be formally allocated to named roles (not individuals by name, but to roles that are then filled by named individuals). An informal understanding within a management team does not satisfy this requirement.

  2. Communication — the assignment shall be communicated within the organisation. This means documented and accessible to the people whose work it governs, not held only in a management decision log.

  3. Reporting — the assigned role shall have a defined channel for reporting AIMS performance back to top management. Without this feedback loop, Clause 5.3(b) cannot be satisfied regardless of other controls in place.

Clause 5.3 cross-references Clause 5.1 (Leadership and Commitment) and Clause 9.3 (Management Review), which together create a closed governance loop: top management sets direction, named roles execute and monitor, and performance data returns to management via a scheduled review. Firms that treat ISO/IEC 42001 as a documentation exercise rather than a governance architecture will fail all three elements.

⚠ Warning — Common Misreading

Clause 5.3 does not permit a "shared ownership" model in which accountability is distributed across the partnership without a single point of coordination. Shared knowledge is not assigned responsibility. The standard requires a named role with documented authority — and that role shall be distinct from the individuals who operate the AI systems being governed.

ITIL 4 Note — Workforce and Talent Management Practice

In ITIL 4 terms, Clause 5.3 maps directly to the Workforce and Talent Management practice, which requires that service roles be defined with clear accountability, authority, and reporting lines. RACI-AIMS-001 functions as the ITIL authority matrix for all AI-related service roles — the instrument that defines who is Responsible, Accountable, Consulted, and Informed for each AI governance task. Without a RACI, AI governance tasks exist in operational practice as unowned service actions: performed without authorisation, tracked without accountability, and unreportable to management.

ITIL 4's Service Configuration Management practice further requires that service assets — including AI systems — shall be assigned an owner. The AI system register obligation in Gap 3 (REG-AIMS-SYS-001) cannot be formally owned and maintained until the RACI role assignment in RACI-AIMS-001 is in place.


The SRA's position: governance is a COLP matter

The SRA's guidance note Compliance Tips for Solicitors Regarding the Use of AI and Technology, updated 9 February 2026, makes the accountability linkage explicit. Firms using AI tools remain responsible for the output under SRA Principles 2 (act with integrity), 5 (properly supervise), and 7 (client best interests). The guidance does not carve out a technology exception.

The SRA's AI guidance, read in conjunction with the December 2025 COLP thematic review, establishes the following implied requirements. Each is framed in the regulatory instrument's own mandatory terms:

SRA Accountability Framework for AI — Regulatory Requirements
01
The COLP shall know what AI tools are in use. SRA Code of Conduct for Firms 2019, Rule 8.1(a) requires the COLP to take all reasonable steps to ensure compliance. This obligation shall extend to AI systems. The thematic review found that most COLPs could not outline their own responsibilities — meaning most firms are unable to demonstrate this step has been taken.
02
Supervision obligations shall extend to AI output. SRA Principle 5 requires proper supervision of work produced within the firm. The SRA has not exempted AI-generated output from this obligation. The COLP's accountability shall include ensuring that supervision mechanisms exist for AI-assisted work product before it is delivered to clients or filed with courts or tribunals.
03
Client care obligations shall be addressed. Where AI tools are used in client matters, the SRA requires firms to consider whether disclosure is appropriate (AI guidance, February 2026). The individual responsible for this decision shall be identifiable and formally designated before a problem arises, not after.
04
Panel and procurement requirements shall be answered by a named role. Large institutional clients are requiring AI governance questionnaires of panel firms. The first question — in a variety of forms — is: who owns AI risk in your firm? A response referencing "the partnership" shall not satisfy a due diligence questionnaire requiring a named role with a documented mandate.
"Only one COLP could outline all their responsibilities."
SRA · Compliance Officers: A Thematic Review · December 2025

The SRA's enforcement posture on AI is developing. The direction is clear from the thematic review: the regulator is stress-testing governance structures, not only outcomes. A firm that cannot demonstrate who owns AI accountability is demonstrating a gap in its compliance architecture — not merely a gap in technology policy.


Why law firms fail Clause 5.3 — the four non-conformance patterns

Across law firm assessments, Clause 5.3 non-conformances cluster into four distinct patterns. Each has a root cause and a remediation path. A firm may exhibit more than one simultaneously.

Cl. 5.3 Non-Conformance Taxonomy
NF1
Role vacuum. No role within the firm has been formally designated as accountable for AI governance. The COLP may hold oversight of regulatory compliance generally, but AI governance is not within scope of any documented mandate. This is the most common and most severe pattern: the firm has no mechanism for ISO/IEC 42001 performance to surface to management.
NF2
Informal designation. Someone — typically the IT lead or a senior partner — handles AI matters in practice, but this has not been documented as a formal role assignment. The gap is not in the person but in the assignment: Clause 5.3 requires the responsibility to be communicated within the organisation, which requires it to exist in a controlled document, not only in institutional memory.
NF3
Incomplete RACI — AI tasks ungoverned. The firm has a RACI for general IT or data governance but has not extended it to AI-specific tasks: reviewing AI system outputs, approving new AI deployments, conducting impact assessments, managing vendor AI contractual obligations, communicating AI use to clients. These tasks exist in operational practice but are unassigned in any governance document — meaning no one is accountable if they are not performed.
NF4
Reporting loop absent. Even where a role exists and a rudimentary RACI is in place, there is no defined mechanism for AI performance data — incidents, near-misses, output quality observations, staff usage patterns — to reach the management board. Without this reporting loop, Clause 5.3(b) cannot be satisfied. The RACI document alone does not close the clause.

The Halstead & Cole scenario

At Halstead & Cole LLP (fictional, SRA number 12345678), 90 fee earners are using three AI systems: GPT-4o, Microsoft 365 Copilot, and LEAP AI. Each was procured separately, by different stakeholders, at different times. No AI governance register exists. No role has been formally assigned accountability for any of them.

Priya Anand, as Partner and COLP, carries statutory accountability for the firm's regulatory compliance — including the supervision obligation under SRA Principle 5. She does not currently know which fee earners are using AI on which matters, whether outputs are being reviewed before filing or sending, or whether Pemberton Capital's panel questionnaire can be answered accurately. She has not been given the tools, the information, or the documented mandate to exercise that oversight.

◈ Caution — Pemberton Capital Deadline

Pemberton Capital's five-question AI governance questionnaire has a 30-day response window. Question 1 is a Clause 5.3 probe: "Who in your firm is responsible for AI governance, and what is their formal mandate?" Without a documented RACI, Halstead & Cole cannot answer this question with confidence — and a hesitant or inconsistent response to a £620,000 per annum client panel relationship carries material commercial risk.

The management resolution in RACI-AIMS-001 Section 1 — a single page, completable within one working day — is sufficient to answer Question 1 within the 30-day window. The full 28-task RACI matrix may follow as supporting evidence.

The second scenario firm — Thornbury Legal Ltd. — illustrates a further variant: the firm operates a self-hosted model for contract analysis and has no RACI that distinguishes supply chain AI governance from internal AI governance. No escalation path exists from the vendor management function to the board when a supply chain AI incident occurs. This variant is addressed in Gap 9 (REG-AIMS-SUP-001) of this series.


Building a compliant AI governance RACI for a law firm

RACI-AIMS-001 is the controlled document that closes this gap. It is not a generic IT governance matrix: it maps AI-specific tasks across the four governance roles appropriate for a law firm of 50–150 fee earners, and it incorporates the reporting loop required by Clause 5.3(b).

The five-step build sequence shall be followed in the order specified:

  1. Designate the AI Governance Lead (AGL). This is the named role accountable to ISO/IEC 42001. In most firms of this scale, this shall be the COLP or a senior partner with a formal mandate. The designation shall be documented in a board or management resolution and cross-referenced in RACI-AIMS-001. The AGL shall be Accountable (A) for all strategic AI governance tasks; they need not be Responsible (R) for operational tasks.

  2. Map the task inventory. Identify every AI-specific task the firm performs or shall perform: AI system procurement approval, usage policy maintenance, impact assessment commissioning, fee earner training delivery, output review sign-off, client disclosure decisions, vendor contract review, incident reporting, and management review presentation. Each task shall appear in the RACI matrix.

  3. Assign RACI across four roles. The four governance roles for a firm of this scale are: AI Governance Lead (strategic accountability), COLP / Practice Manager (regulatory interface), IT Lead / Practice Technology Manager (system-level operations), and Fee Earner / Supervisor (point-of-use accountability). For each task, one role shall be Accountable (A), one or more shall be Responsible (R), and others shall be Consulted (C) or Informed (I) as appropriate. No task shall have more than one Accountable role.

  4. Build the reporting loop. Define a quarterly AI performance report — cadence, format, presenter, and audience — within RACI-AIMS-001 itself. This is the mechanism that satisfies Clause 5.3(b). The report shall cover: AI systems in active use, any incidents or near-misses, output quality observations, training compliance, and any supply chain changes. Without this element, the RACI does not close the clause.

  5. Issue and communicate as a controlled document. Controlled documents under ISO/IEC 42001:2023 Clause 7.5 shall be version-controlled, approved, distributed, and accessible. RACI-AIMS-001 shall be circulated to all role-holders with a read-receipt mechanism documented. This step is not administrative housekeeping — it is the communication step required by Clause 5.3 itself. The role assignment is not complete until it has been communicated to the individuals holding those roles.

ℹ Note — RACI vs. Role Description

RACI-AIMS-001 operates at the task level, not the role description level. It does not replace job descriptions or partnership agreements — it supplements them with AI-specific task assignments. Where a task creates a new obligation for an existing role (e.g., adding AI output review to a supervisory partner's responsibilities), this should be reflected in their role description at the next annual review, but the RACI is the primary governance instrument for ISO/IEC 42001:2023 Clause 5.3 purposes.

Preview: RACI-AIMS-001 core task matrix

Task AI Gov. Lead COLP / PM IT Lead Fee Earner / Supervisor
AI system procurement approvalACRI
AI usage policy maintenanceARCI
AI risk & impact assessmentARCC
Fee earner AI training deliveryACRI
AI output review (matter-level)IAIR
Client AI disclosure decisionCAIR
AI vendor contract reviewACRI
AI incident reportingARCR
Quarterly AI performance report to managementRCCI
RACI annual review & reissueARCI

R = Responsible · A = Accountable · C = Consulted · I = Informed · Rule: one A per task · Full 28-task matrix in RACI-AIMS-001


RACI-AIMS-001: AI Governance Roles, Responsibilities & Authorities

The controlled document that closes Gap 2 is available for download below. It includes the full 28-task RACI matrix across four governance roles, a management resolution template for the AGL designation, a quarterly reporting schedule, and the version control and communication log required by ISO/IEC 42001:2023 Clause 7.5.

RACI-AIMS-001 v1.0 · ISO/IEC 42001:2023 Cl. 5.3 · Word Document (.docx) · No email required
AI Governance Roles, Responsibilities & Authorities

28-task RACI matrix · 4 governance roles · Management resolution template · Quarterly reporting schedule · Version control log · Communication register. Editable Word format — adapt role names to your firm's structure, obtain management sign-off, and issue as a controlled document under your AIMS.

How to use RACI-AIMS-001 within 30 days: Begin with the management resolution in Section 1 — this formalises the AGL designation and is the fastest single action available to close the Clause 5.3 role vacuum. Then work through the task matrix in Section 2, confirming or adjusting the default role assignments for your firm's structure. Issue the completed document under your document control system and circulate with a read-receipt. The communication step shall be completed and recorded before the RACI is considered active.

◈ Caution — Scope of the RACI

RACI-AIMS-001 covers internal AI governance responsibilities. It does not address third-party and supply chain AI governance, which is handled separately in Gap 9 (REG-AIMS-SUP-001). If your firm uses external AI vendors or self-hosted models procured from third parties — as Thornbury Legal Ltd. does — both documents are required. The vendor management tasks in RACI-AIMS-001 represent the internal accountability dimension; REG-AIMS-SUP-001 addresses the supplier-facing controls.


Where Gap 2 sits in the remediation programme

Clause 5.3 non-conformance is classified as Critical in the Halstead & Cole gap register because it is a prerequisite for every subsequent gap. Without a documented AI governance RACI, there is no role accountable for completing the gap register, commissioning risk assessments, maintaining the AI system register, or reviewing supply chain controls. Gap 2 is not just a governance gap — it is the governance gap.

In the 90-day remediation roadmap, closing Gap 2 is a Week 1 action within Phase 1 (Foundation). It runs in parallel with Gap 1 (AI Policy, POL-AIMS-001) and shall precede Gap 3 (AI System Register, REG-AIMS-SYS-001), which requires an accountable role to exist before the register can be owned and maintained as a controlled document.

Remediation Sequence — G2 Dependencies
←
Predecessor: G1 (AI Policy). POL-AIMS-001 shall be in draft before RACI-AIMS-001 is issued, because the policy establishes the governance framework within which the RACI operates. The AGL designation in the RACI shall cross-reference the policy by document ID.
→
Successor: G3 (AI System Register). REG-AIMS-SYS-001 requires an owner. The RACI shall identify which role is Accountable for the register before the register can be formally issued as a controlled document. Before you can govern AI, you shall know what AI you have.
→
Successor: G4 (Risk & Impact Assessment). The RACI's "AI risk & impact assessment" task row determines who commissions and approves assessments. Without this assignment, the Gap 4 remediation process has no defined owner and cannot proceed.
⊕
Parallel: Pemberton Capital response. The RACI-AIMS-001 management resolution — Section 1, one page — is sufficient to answer Pemberton Capital's Question 1 ("Who owns AI risk in your firm?") within the 30-day response window. The full 28-task matrix may follow as supporting evidence.

The next article in this series covers Gap 3: the AI system register and prompt version control obligations under ISO/IEC 42001:2023 Clause 7.5. Gap 3 — AI System Register & Prompt Version Control →


Quality gate record

The following gates were checked before publication of v1.1. All 10 gates pass. This record is maintained as evidence of the pre-publication compliance audit and satisfies ISO 9001:2015 Clause 7.5.3 (control of documented information) for the article series.

Close Gap 2 — download RACI-AIMS-001

The full 28-task AI governance RACI matrix, management resolution template, and quarterly reporting schedule. Editable Word document, ready for management sign-off. No email required.

Download RACI-AIMS-001 (.docx) → View full gap register

This article is published for educational purposes only and does not constitute legal advice or advice on compliance with ISO/IEC 42001:2023 or any other standard. Halstead & Cole LLP, Thornbury Legal Ltd., Priya Anand, Pemberton Capital LLP, and all associated names are fictional constructs used for illustrative purposes. Any resemblance to real persons or firms is coincidental. All templates and controlled document examples require adaptation and qualified review before use. · Document ref: ART-AIMS-G002 v1.1 · Published 3 July 2026 · Amended 4 July 2026 · Next review: 3 October 2026 · Retention: 7 years · UNUS London Ltd. · unuslondon.com/legal/iso-42001-gap-2-colp-roles