AI leadership appointment as a governance problem for Australian boards
Across Australian and global organisations, AI leadership roles (for example, “Head of AI” or “Chief AI Officer”) are being created and filled at speed, often through processes that mis-specify the role, mis-constitute the selection panel, and misalign assessment criteria with the actual governance challenge. The result is a systematic pattern of appointments that are technically credible yet strategically and governance-inadequate, compounding rather than mitigating AI-related risk.
This paper advances the theoretical claim that AI leadership appointments fail systematically because organisations apply industrial-era change frameworks to a post-industrial governance problem, and the panels making the decision lack the competence to recognise the mismatch. Drawing on four literature, emerging work on AI leadership and the Chief AI Officer, corporate governance of AI (including ASIC Report 798, APRA CPS 230, Corporations Act 2001 (Cth) s 180, and leading board-level guidance), sociotechnical systems theory and organisational frameworks, and responsible AI practitioner research, the paper reframes the AI leadership hire as a board-level governance decision rather than a technology procurement.
It offers an AI Leadership Appointment Readiness (ALAR) Framework that synthesises three recurrent failure modes, role mis-specification, panel dysfunction, and criteria misalignment, into a diagnostic model anchored in Australian directors’ duties, prudential expectations, and AI governance standards. The paper concludes with practical implications for Australian boards seeking to discharge their duty of care and diligence in appointing AI leaders who can deliver commercially sound, ethically grounded, and regulator-ready AI capability.
Keywords: AI governance, AI leadership, Chief AI Officer, directors’ duties, appointment process, sociotechnical systems, ASIC Report 798, APRA CPS 230
1. Introduction
For many boards, the decision to appoint an AI leader will be the most consequential organisational appointment they make in the next five years, shaping how their organisations sense, decide, and act in an AI-saturated environment. Yet evidence from practice suggests that boards and executive teams are frequently getting the hire wrong, not occasionally, but systematically.
AI leadership appointments are routinely made using job descriptions recycled from digital transformation or ICT roles, overseen by panels without AI governance expertise, and evaluated through interviews that test tool familiarity rather than strategic, ethical, and risk-management capability. This pattern produces a leadership gap that amplifies every other AI risk the organisation faces, from model failure and bias to operational disruption and regulatory exposure.
This paper proceeds from a central theoretical claim: AI leadership appointments fail systematically because organisations apply industrial-era change frameworks to a post-industrial governance problem, and the panels making the decision lack the competence to recognise the mismatch. Boards and executives are using frameworks designed for stable, tool-centric change (people–process–technology (PPT) thinking, ICT project governance) to address a sociotechnical governance challenge in which agency is distributed across humans, models, infrastructure, vendors, and regulators in ways that evolve continuously and cannot be fully anticipated.
In the Australian context, this failure is not merely sub-optimal; it sits directly against an emerging regulatory backdrop in which ASIC, APRA, and the Australian Institute of Company Directors (AICD) are explicit that boards retain oversight responsibility for AI risk, governance, and operational resilience. ASIC Report 798 explicitly links AI use to existing directors’ duties under the Corporations Act 2001 (Cth), including the duty of care and diligence in s 180. APRA’s CPS 230, effective from 2025, reinforces board accountability for operational risk management, including AI-enabled systems and third-party arrangements.
Against this backdrop, appointing an AI leader is not a routine technology hire; it is a governance decision with direct implications for directors’ duties, prudential expectations, and the organisation’s social licence. This paper offers a diagnostic framework (the AI Leadership Appointment Readiness (ALAR) Framework) to help boards assess and improve the governance quality of their AI leadership appointments.
The paper proceeds as follows. Section 2 reviews four relevant bodies of literature and identifies the gap each leaves with respect to AI leadership appointment. Section 3 proposes the ALAR Framework and diagnoses three recurrent failure modes. Section 4 examines why industrial-era frameworks are theoretically inadequate for this class of problem. Section 5 offers a normative governance framework for AI leadership appointment grounded in Australian regulatory expectations. Section 6 addresses implications for Australian boards, focusing on regulatory and competitive stakes. Section 7 concludes with directions for further research.
2. Literature Review
2.1 AI Leadership Roles and the Chief AI Officer
Emerging scholarship on AI leadership highlights the need for dedicated executive roles to integrate AI strategy, governance, and implementation at enterprise scale. Schmitt’s (2024) analysis of the Chief AI Officer (CAIO) argues that existing C-suite roles (the CIO, CDO, and CTO) struggle to resolve the structural tensions introduced by AI: between innovation and risk, automation and human judgment, and centralisation and decentralisation. A CAIO, Schmitt argues, can provide a dedicated focal point for AI strategy and governance, bridging technology, operations, risk, and ethics across the enterprise.
Related work emphasises that AI’s value is realised not through isolated tools but through re-architecting decision processes and business models (Lonsdale, 2026). This positions AI leadership beyond “innovation theatre” or vendor management toward a sustained executive function with strategic accountability. The Institute of Managers and Leaders (IMIA, 2026) further notes that AI is reshaping the strategic leadership mandate in ways that require new competencies at the executive and board level.
This literature establishes a compelling rationale for dedicated AI leadership and begins to sketch the capability profile such leadership should possess. However, it does not examine how these roles are appointed in practice, or whether existing appointment processes are capable of identifying the capabilities they describe. The question of who selects AI leaders, using what criteria and what panel composition, remains largely unaddressed. This is the gap the present paper addresses.
2.2 Corporate Governance of AI
A growing body of governance guidance frames AI explicitly as a board-level responsibility. Gregory (2023), writing for the Harvard Law School Forum on Corporate Governance, sets out a three-part approach to AI oversight: understanding AI’s role in strategy, ensuring risk and control frameworks are updated, and embedding AI considerations into board committees. The National Association of Corporate Directors (NACD, 2025) similarly urges boards to align AI policy and oversight with organisational values, rather than treating AI as a siloed technology issue.
In Australia, the AICD and Human Technology Institute’s (2024) Director’s Guide to AI Governance outlines eight elements of safe and responsible AI governance, emphasising that boards must determine appropriate governance structures, clarify accountability, and leverage external expertise where necessary. ASIC’s Report 798 (2024) identifies significant variation in AI governance maturity among Australian financial services licensees, finding weaknesses that create “potential for gaps as AI use accelerates” and explicitly linking AI governance to directors’ duties under the Corporations Act 2001 (Cth) s 180. APRA’s CPS 230 on Operational Risk Management formalises board responsibility for operational resilience, including risks arising from AI, outsourcing, and third-party providers (Clifford Chance, 2025; Canon Business, 2025). Commentary from MinterEllison (2024) and Hall and Wilcox (2026) confirms that ASIC expects governance arrangements to lead rather than lag AI use.
Collectively, this literature situates AI firmly within the remit of board oversight and reinforces that AI governance is a directorial responsibility, not merely a management function. However, it does not address the specific appointment process through which the AI leaders responsible for delivering that oversight are selected. A robust governance framework for AI oversight is directly undermined if the person appointed to operationalise it is selected through a process that cannot assess the relevant capabilities. This is the procedural gap the ALAR Framework addresses.
2.3 Organisational Frameworks and Their Limits
Traditional digital and ICT change programmes rely on PPT frameworks and project-centric governance models that assume a relatively stable technology artefact and linear implementation phases. Sociotechnical systems theory and sociomaterial perspectives challenge these assumptions fundamentally.
Orlikowski (2007) argues that technology and organisational practice are mutually constitutive: they co-evolve through ongoing enactment rather than through one-off implementation events. Leonardi (2012) develops this through the concept of imbrication, showing how human and material agencies become interwoven over time in ways that cannot be reduced to either “the system” or “the organisation” as discrete, manageable entities. Hutchins’s (1995) work on distributed cognition further demonstrates that knowledge, judgment, and decision-making capability are distributed across human actors, artefacts, and their configurations, not concentrated in a single person, team, or system.
Applied to AI, these perspectives carry significant implications. AI systems are not stable artefacts that can be installed and then left to operate; they are adaptive systems whose behaviour evolves with data, user interaction, and environmental change. Reviews of AI governance literature confirm that AI introduces opacity, feedback loops, and emergent behaviour that strain conventional control paradigms (Rismani & Moon, 2023). Governance must therefore be continuous, adaptive, and designed for distribution.
Although sociotechnical theory has been applied to AI governance broadly, it has not previously been applied to the specific context of AI leadership appointment, where framework assumptions shape not just implementation but candidate selection and role design. PPT and ICT frameworks applied to appointment processes systematically privilege candidates experienced in tool deployment and project delivery over those capable of designing adaptive, distributed governance systems. This paper applies sociotechnical theory to that specific and consequential context.
2.4 Responsible AI Practitioner Literature
Rismani and Moon (2023) develop an ontology of responsible AI practitioner roles and skills, demonstrating that responsible AI work is inherently multi-disciplinary, spanning technical, legal, design, and organisational domains. They identify practitioner functions, risk assessment, stakeholder engagement, impact analysis, and governance design, that sit well beyond the scope of traditional engineering or ICT roles. Complementary work emphasises that sustainable AI transformation requires leadership capable of managing human–AI integration at a cultural level, not merely an operational one (Potential Project, n.d.; Ibarra & Wilkinson, 2026).
This literature provides a rich account of what effective AI practitioners do and what capabilities they need to do it. It implies a meaningful benchmark for what AI leadership selection should assess. However, it has not been operationalised into appointment frameworks or assessment criteria that boards can apply directly. The ALAR Framework attempts to bridge this gap, translating the practitioner capability literature into a board-usable diagnostic for appointment design.
3. The Appointment Problem: An AI Leadership Appointment Readiness (ALAR) Framework
Drawing together practice observations and the four literatures reviewed above, this section proposes the ALAR Framework as a diagnostic for boards and executives. The framework identifies three recurrent failure modes in AI leadership appointment (role mis-specification, panel dysfunction, and criteria misalignment) and conceptualises appointment readiness along three corresponding axes.
3.1 Role Mis-Specification
The first failure mode is role mis-specification: organisations recycle descriptions from “Head of Digital”, “Head of Data”, or “Head of ICT” profiles, append AI terminology, and assume the exercise is complete. These legacy role designs implicitly frame AI as a technology problem, privileging technical credentials over governance, risk, and organisational leadership capability. As the corporate governance literature makes clear, AI governance requires a fundamentally different capability profile from ICT management (AICD & Human Technology Institute, 2024; ASIC, 2024).
Mis-specification tends to produce two archetypal and inadequate hires. The “technologist hire” places a competent engineer or data scientist into a role that actually requires enterprise change leadership, regulatory fluency, and board-level communication. The “evangelist hire” appoints a charismatic storyteller with strong AI literacy but insufficient depth in risk, compliance, and implementation discipline. Both profiles can pass interviews built around tool familiarity and generic leadership questions; neither is adequate for the governance-centric role the organisation actually needs.
3.2 Panel Dysfunction
The second failure mode is panel dysfunction. Typical AI leadership hiring panels consist of HR generalists, ICT managers, and business unit representatives, with no AI governance expertise in the room. This is a structural design failure, not a criticism of the individuals involved. Panel composition directly determines the questions asked and the criteria applied; a panel without AI governance expertise will inevitably default to what it already knows.
In practice, this produces interviews dominated by technical trivia (“Which large language models have you used?”), tool familiarity questions (“What AI platforms have you implemented?”), and generic leadership prompts (“Tell us about a time you led a team through change.”). None of these questions tests whether candidates can integrate AI with strategy, risk, culture, and ethics at enterprise scale. Consistent with Rismani and Moon’s (2023) finding that responsible AI work spans technical, legal, design, and organisational domains, effective assessment requires panel members capable of evaluating candidates across all of those dimensions, not just the technical one.
3.3 Criteria Misalignment
The third failure mode is criteria misalignment: role success is implicitly defined in terms of tool procurement and deployment rather than governance outcomes. Job descriptions frequently list “selecting and implementing AI platforms” and “managing AI vendor relationships” as primary accountabilities, framing AI leadership as a procurement and IT delivery function.
The governance literature is clear that this framing is inadequate. ASIC (2024) calls for governance arrangements that “lead” AI use rather than lag it; AICD and Human Technology Institute (2024) frame AI governance as encompassing risk management, ethics, accountability, and organisational capability, not vendor selection. When assessment criteria are misaligned with these governance expectations, even well-intentioned panels will select candidates optimised for the wrong problem.
3.4 The ALAR Framework
The ALAR Framework conceptualises AI leadership appointment readiness along three axes, mandate clarity, panel capability, and criteria alignment, each corresponding to one of the failure modes identified above. Table 1 provides a diagnostic across three readiness levels for each dimension.
Table 1: AI Leadership Appointment Readiness (ALAR) Framework Diagnostic
| Dimension | Low Readiness | Moderate Readiness | High Readiness |
|---|---|---|---|
| Mandate Clarity | Role recycled from ICT or digital; no strategic or governance problem explicitly articulated; success defined by tool deployment. | Role partially redesigned; governance elements referenced but not operationalised; success criteria partially defined. | Mandate explicitly anchors the role in strategic, risk, and governance outcomes; success measured in value, risk reduction, and capability; board-reviewed. |
| Panel Capability | HR generalist + ICT manager + business unit rep only; no AI or governance expertise present. | AI-adjacent expertise (data science, digital transformation) present; no independent AI governance expertise. | Board or committee rep + executive sponsors + technology and data leadership + independent AI governance expertise. |
| Criteria Alignment | Assessment dominated by tool familiarity, technical trivia, and generic leadership questions; governance absent. | Mix of technical and leadership criteria; governance and ethics partially addressed; responsible AI largely absent. | Assessment anchored in strategic integration, governance fluency, regulatory literacy, change leadership, and responsible innovation. |
Organisations with low readiness across all three dimensions are structurally predisposed to mis-hire, regardless of the quality of the candidate pool. Moderate readiness organisations have partially addressed the problem, typically by redesigning the role description or adding a technology-adjacent panel member, but remain exposed on the dimensions they have not addressed. High readiness organisations treat the appointment as a board-level governance decision, design the mandate accordingly, assemble panels capable of evaluating what actually matters, and anchor assessment in strategic, governance, and change leadership capability.
The ALAR Framework is intended as a practical diagnostic, not a linear maturity model. Boards may be strong on mandate clarity but weak on panel capability; the framework surfaces these gaps independently so they can be addressed specifically and sequentially rather than treated as a single undifferentiated problem.
4. Why Existing Frameworks Fail: A Theoretical Account
Industrial-era change frameworks, including PPT, ICT project governance, and standard digital transformation models, were designed for a different class of problem: relatively stable technologies implemented into largely unchanged organisational architectures, with clear boundaries between “the system” and “the organisation”, and linear project phases from design through to operation. Agentic and adaptive AI systems challenge these assumptions in at least four fundamental ways.
First, opacity and drift. Machine learning systems can evolve in ways that are difficult to inspect post-deployment, requiring continuous monitoring and recalibration rather than one-off implementation and training. The assumption of a stable, inspectable artefact does not hold for AI.
Second, feedback loops. AI outputs shape human behaviour and organisational processes, which in turn alter the data on which future model behaviour is based, creating dynamic feedback loops that PPT frameworks have no mechanism to account for or govern.
Third, distributed agency. Responsibility for AI outcomes is spread across developers, operators, data providers, infrastructure vendors, regulators, and end-users, in patterns that shift over time as systems and their uses evolve. Orlikowski (2007) argues that agency in technology-infused work is not concentrated in discrete actors or artefacts but emerges from their ongoing mutual constitution. Hutchins’s (1995) framework of distributed cognition extends this insight: decision-making capability and accountability are distributed across networks of human and non-human actors, and governance must be designed for that distribution rather than assuming it can be resolved into a single accountable agent. Conventional principal–agent accountability models are structurally inadequate for this context.
Fourth, regulatory entanglement. AI use engages overlapping regulatory regimes, corporations law, privacy, anti-discrimination, prudential standards, and emerging AI-specific regulation, making governance a multi-jurisdictional exercise rather than an internal process design question.
PPT framings, centred on “aligning people and processes around tools”, are unable to account for these dynamics. They reinforce a conception of AI as an instrument to be installed rather than a sociotechnical system to be continuously governed. Leonardi’s (2012) concept of imbrication is instructive here: organisational and material agencies become interwoven over time in ways that cannot be undone or managed through one-off alignment exercises. Governing AI requires ongoing attention to how human practices and AI behaviours co-evolve, precisely the kind of sustained, adaptive oversight that PPT frameworks are not designed to support.
In appointment processes, this theoretical inadequacy becomes directly consequential. Panels applying PPT-derived criteria seek candidates with experience in tool deployment and process change management. Candidates with expertise in adaptive governance design, regulatory navigation, and distributed accountability, the skills the literature reviewed in Section 2 identifies as essential, are systematically undervalued by those criteria. The mismatch is not incidental; it is a predictable consequence of applying a theoretically inadequate framework to the wrong class of governance problem.
5. A Governance Framework for AI Leadership Appointment
If existing frameworks mis-specify the problem, what should a governance-aligned appointment process look like within Australia’s regulatory context? Building on ASIC Report 798, APRA CPS 230, the Corporations Act 2001 (Cth), and AICD guidance, this section proposes three normative design principles for AI leadership appointment.
5.1 Mandate Design Anchored in Directors’ Duties
Directors’ duties under s 180 of the Corporations Act 2001 (Cth) require that powers be exercised with the care and diligence of a reasonable person in the same position and circumstances. ASIC (2024) has signalled that these duties extend to AI adoption and governance, urging directors to understand how AI is used, the extent to which they rely on AI-generated information, and the associated risks and controls. APRA’s CPS 230 reinforces this by requiring boards to take responsibility for the entity’s operational risk management framework, including risks arising from AI and third-party providers (Clifford Chance, 2025).
A governance-aligned mandate for an AI leader should therefore explicitly state: the strategic, risk, and governance problems the role is appointed to solve; the decisions the role will own, influence, or inform at executive and board level; and how success will be measured in terms of value creation, risk reduction, and capability building over 12–36 months. Mandate design should be reviewed by the board’s risk committee or equivalent to ensure alignment with directors’ duties, prudential expectations, and enterprise risk appetite. A mandate that cannot be articulated in these terms is not ready to be filled.
5.2 Panel Composition Aligned with Governance Expectations
AICD and Human Technology Institute (2024) recommend that boards determine which committees will oversee AI and consider how external experts can be engaged in governance. This principle applies directly to appointment panels. A panel that lacks AI governance expertise cannot assess AI governance capability, which, as this paper has argued, is the primary capability the role requires.
A governance-aligned panel should include: board representation (from the risk or people and culture committee, to anchor selection in directors’ duties and strategic priorities); executive sponsorship (CEO, COO, CRO, or CDO, reflecting the enterprise scope of the role); technology and data leadership (CIO, CISO, or Head of Data, to assess technical literacy and integration complexity); and independent AI governance expertise, external practitioners whose primary work is advising boards and executives on AI risk, governance, ethics, and compliance. The last element is not optional. Without it, the panel defaults to the ICT and HR framings that produce the failure modes documented in Section 3, regardless of the quality of the other panellists.
5.3 Assessment Criteria Aligned with AI Governance Maturity
Drawing on the AICD’s eight elements of AI governance, ASIC’s diagnostic questions in Report 798, and Rismani and Moon’s (2023) practitioner capability ontology, assessment criteria should encompass five capability domains: strategic and commercial integration (linking AI to business value and competitive positioning, not merely experimentation); governance, ethics, and regulatory fluency (navigating privacy, prudential, anti-discrimination, and AI-specific obligations); technical literacy sufficient to interrogate vendors and challenge assumptions, without conflating leadership with engineering; change leadership and culture-building at enterprise scale; and responsible innovation and assurance, including the judgment to stop or reshape initiatives when risk outweighs benefit.
Assessment should use behavioural questions that probe judgment and trade-offs rather than technical inventories or experience checklists. Asking candidates to describe a time they stopped or reshaped an AI initiative due to unacceptable risk, or to explain how they would present AI monitoring data to a board in terms the board can act on, reveals far more about governance capability than asking which tools they have used or which AI certifications they hold.
6. Implications for Australian Boards
The convergence of AI adoption acceleration, evolving regulatory expectations, and increasing stakeholder scrutiny creates both obligation and material risk for Australian boards. The stakes of getting the AI leadership appointment wrong are not abstract, and they compound over time.
6.1 Regulatory Exposure
ASIC’s finding in Report 798 that AI governance arrangements “varied widely” and often exhibited weaknesses is a direct signal of regulatory attention, not a historical observation. Where AI governance failures are traceable to inadequate leadership, including leadership that was inadequately appointed, ASIC has a clear basis to examine whether directors discharged their s.180 duties. Commentary from MinterEllison (2024) and Hall and Wilcox (2026) confirms that ASIC expects licensees to treat Report 798 as a prompt for corrective action, not a baseline to be monitored.
CPS 230’s operational risk provisions, effective from 2025, add further weight. Boards that cannot demonstrate their AI leaders were appointed through a process aligned with governance expectations will find it increasingly difficult to satisfy prudential supervisors that their operational risk management frameworks are adequate. The appointment process is itself a governance artefact and is therefore subject to regulatory scrutiny in its own right.
6.2 The Compounding Cost of a Mis-Hire
Beyond regulatory exposure, the strategic cost of a mis-hire compounds over time in ways that are difficult to reverse. An AI leader appointed to manage tool procurement will build a function oriented around tool management. The organisation’s AI capability will develop in the wrong direction, optimised for vendor relationships rather than governance maturity, for deployment metrics rather than value creation and risk control. Organisational capital, budget, talent, attention, and credibility will be consumed by a function solving the wrong problem.
By the time the gap becomes visible, typically through an AI-related incident, a failed initiative, or a regulatory enquiry, the organisation will have made significant technology, process, and people investments that are difficult to unwind. The true cost of the mis-hire is therefore not the appointment itself; it is the opportunity cost of the governance capability the organisation failed to build, plus the remediation cost of the investments made in the wrong direction.
6.3 The Appointment as Governance Signal
Conversely, boards that treat AI leadership appointment as a governance decision, designing the mandate carefully, assembling a capable panel, applying criteria aligned with governance maturity, send a signal both internally and externally that AI is being taken seriously at the board level. This matters for talent attraction: the best AI governance practitioners actively seek organisations that understand what they do and give them the mandate and reporting lines to do it well. It matters for regulatory relationships: proactive, documented governance is a meaningful differentiator when regulators are assessing governance maturity. And it matters for investor and stakeholder confidence in an environment where AI governance failures are increasingly visible and consequential.
For sectors beyond financial services, where CPS 230 does not apply directly, the logic holds equally. Directors’ duties, stakeholder expectations, and emerging AI-specific regulation are converging on similar governance requirements across sectors and jurisdictions. The AI leadership appointment will become progressively harder to defend, to regulators, investors, and the community, as scrutiny intensifies and the bar for demonstrable governance maturity rises.
7. Conclusion
AI is not a tool that can be installed through conventional ICT change programmes. It is a structural shift in how organisations sense, decide, and act, with governance implications that reach the boardroom and engage directors’ duties, prudential obligations, and social licence. This paper has argued that systematic failures in AI leadership appointment arise because organisations apply industrial-era change frameworks to a post-industrial sociotechnical governance problem, and because appointment panels lack the competence to recognise the mismatch.
By situating AI leadership within four bodies of literature, CAIO and AI leadership roles, corporate governance of AI, sociotechnical systems theory, and responsible AI practitioner research, and anchoring the analysis in Australian regulatory context (ASIC Report 798, Corporations Act 2001 (Cth) s 180, APRA CPS 230, AICD guidance), this paper reframes the AI leadership appointment as a core governance decision, not a specialised HR task.
The ALAR Framework provides a diagnostic for boards to assess their own appointment readiness across three dimensions: mandate clarity, panel capability, and criteria alignment. The normative governance framework proposed in Section 5 offers practical design principles grounded in Australian directors’ duties and prudential expectations. Together, they provide boards with a starting point for redesigning AI leadership appointment in a way that is governance-sound, practically achievable, and demonstrably aligned with regulatory expectations.
Further research opportunities include empirical studies of AI leadership appointment processes across Australian sectors, examining actual panel composition, question design, and selection outcomes; analysis of board committee charters and minutes to understand how AI oversight is operationalised in practice; and longitudinal evaluation of how appointment strategy affects AI governance maturity, risk incidents, and regulatory interactions over time. The present analysis is necessarily conceptual and diagnostic; empirical validation of the ALAR Framework across sectors and organisational types would significantly strengthen its practical utility and theoretical standing.
The practical implication remains clear. Boards that treat AI leadership appointment as a peripheral technology hire are not managing AI risk; they are creating it. Boards that redesign the mandate, panel, and criteria through a governance lens will be better placed to appoint AI leaders capable of delivering commercially sound, ethically grounded, and regulator-ready AI capability, and to demonstrate to ASIC, APRA, and their stakeholders that they have done so.