<< Back

ISO/IEC 38507: Governance Implications of AI for Organizational Directors

Best Practices / Lessons Learned

Why boards must move beyond AI adoption and take ownership of AI accountability, organizational value, and responsible innovation

An organization can invest millions in artificial intelligence, deploy sophisticated AI models, automate critical business processes, and still fail at one fundamental responsibility: governing how AI is used and what happens as a result.

 

Imagine a board approving an AI-powered recruitment platform to improve hiring efficiency. Six months later, the organization discovers that the system systematically disadvantages certain candidates.

Who approved the business purpose? Who evaluated the potential discrimination? Who monitored the outcomes? Who was responsible for intervening?

These are not merely technical questions. They are questions of organizational governance.

This is precisely why ISO/IEC 38507:2022 — Information technology — Governance of IT — Governance implications of the use of artificial intelligence by organizations deserves the attention of directors, board members, executive leaders, Chief Information Officers, Chief Risk Officers, AI leaders, and project management professionals.

The standard provides guidance to members of an organization's governing body on enabling and governing AI to ensure its effective, efficient, and acceptable use.

Its message is particularly important in today's environment: AI may operate at machine speed, but the responsibility for governing its use remains with people.

  1. What is ISO/IEC 38507, and why should directors care?

Published in April 2022, ISO/IEC 38507 extends the conversation around IT governance into the opportunities, risks, and organizational responsibilities associated with AI.

It is intended for governing bodies, but its relevance extends to executive managers, technical and legal specialists, service providers, auditors, assessors, and policymakers.

Unlike a technical AI standard that focuses on how a model is developed or tested, ISO/IEC 38507 focuses on the governance implications of an organization's use of AI.

It encourages governing bodies to understand:

  • The opportunities AI creates for organizational value and innovation.
  • The risks and responsibilities introduced by AI adoption.
  • The implications of automation for authority, responsibility, and accountability.
  • The impact of AI decisions on customers, employees, partners, and other stakeholders.
  • The need to align AI use with organizational objectives, policies, ethics, and legal obligations.
  • The importance of appropriate oversight throughout the use of AI.

An important distinction: ISO/IEC 38507 is governance guidance, not a technical model-development specification or a standalone AI management-system certification checklist.

Why this matters: Directors do not need to become data scientists or understand every algorithm. They do need to understand what AI is being used for, what risks it introduces, who has authority over it, and how the organization knows that its use remains acceptable.

  1. The boardroom challenge: AI decisions are becoming organizational decisions

Traditionally, technology decisions were often delegated to IT departments. AI changes the nature and scale of those decisions.

The board's responsibility is therefore not simply to approve AI investments. It must ensure that the organization has the governance capabilities to oversee their use.

The critical shift: AI governance is not just an IT responsibility. It is an organizational leadership responsibility.

  1. Real-world cases: What happens when AI governance is overlooked?

The following cases illustrate why the governance questions addressed by ISO/IEC 38507 matter in practice.

Case 1: Amazon's experimental AI recruitment tool — When historical data reinforces bias

In 2018, Reuters reported that Amazon had discontinued an experimental AI recruitment tool after discovering that it disadvantaged women.

The system had been trained using historical resumes, many of which came from men, reflecting the gender imbalance in the technology industry. It learned patterns that could disadvantage resumes containing terms associated with women and graduates of certain women's colleges.

The company reportedly attempted to address some of these issues but ultimately abandoned the project.

Governance lesson for directors:

A technically sophisticated system can reproduce historical organizational inequalities.

Leadership teams should ask:

  • Was the training data representative?
  • Were potential discriminatory outcomes assessed?
  • Were affected stakeholder groups considered?
  • Was there a process for independent validation?
  • Could the organization identify and stop unacceptable outcomes?

Project management takeaway: Include fairness testing, data-quality reviews, stakeholder impact assessments, and explicit acceptance criteria in the AI project lifecycle—not as activities postponed until after deployment.

Case 2: Air Canada chatbot — An organization remains responsible for its AI

In Moffatt v. Air Canada (2024 BCCRT 149), a passenger relied on incorrect information from Air Canada's website chatbot regarding bereavement fares.

The chatbot indicated that a passenger could apply for a reduced fare after travel within a specified period. The airline's actual policy required the request to be made before travel.

When the passenger sought the reduction, the airline refused.

The British Columbia Civil Resolution Tribunal found Air Canada responsible for the misleading information and awarded compensation.

The case became an important reminder that organizations cannot simply distance themselves from inaccurate information generated by AI systems they deploy.

Governance lesson for directors:

An AI chatbot is part of the organization's customer-facing operation. Its outputs can create financial, legal, and reputational consequences.

Directors should expect management to establish:

  • Clear ownership of customer-facing AI systems.
  • Testing against approved organizational policies.
  • Monitoring of inaccurate or harmful responses.
  • Complaint and escalation procedures.
  • Human intervention for sensitive or high-impact situations.

Project management takeaway: AI acceptance testing must include realistic customer scenarios, policy changes, exception handling, and ongoing production monitoring.

Case 3: NIST AI Risk Management Framework — Governance must continue throughout the lifecycle

The U.S. National Institute of Standards and Technology (NIST) introduced its AI Risk Management Framework (AI RMF 1.0) to help organizations manage risks associated with AI.

Its four core functions are:

  • GOVERN: Establish organizational policies, responsibilities, and risk-management culture.
  • MAP: Understand the context in which AI operates and identify associated risks.
  • MEASURE: Assess, analyze, and track identified risks.
  • MANAGE: Prioritize and respond to risks throughout the AI lifecycle.

Governance is designed as a cross-cutting function that informs the other three.

Although NIST AI RMF is not ISO/IEC 38507, it offers a useful operational companion for organizations translating board-level governance expectations into risk-management activities.

Governance lesson for directors:

AI oversight cannot be limited to an approval meeting before launch. It must continue as the system, its data, users, business environment, and risks evolve.

Project management takeaway: Treat AI governance as a continuous delivery and operational discipline, supported by measurable controls and evidence.

  1. What should organizational directors actually govern?

ISO/IEC 38507 encourages governing bodies to understand the implications of AI use and ensure that appropriate governance arrangements exist.

For practical leadership discussions, these implications can be organized into six areas.

  1. Strategic alignment and organizational value

AI adoption should support the organization's purpose, strategy, and measurable objectives.

Directors should ask:

  • What business or societal problem are we solving?
  • Is AI genuinely the appropriate solution?
  • What measurable value will it create?
  • Are we accepting risks that are disproportionate to the expected benefits?

An organization should not adopt AI simply because competitors are doing so.

  1. Authority, responsibility, and accountability

Automation can change how decisions are made, but it must not create uncertainty about who is responsible.

Directors should ensure that:

  • AI decision-making authority is clearly defined.
  • Management responsibilities are assigned.
  • Human intervention and escalation mechanisms exist.
  • High-impact decisions have appropriate review and approval.
  • Third-party AI providers do not create accountability gaps.

A useful boardroom question is:

If an AI system makes a consequential decision tomorrow, can we identify the person or function accountable for overseeing that decision today?

  1. Risk, compliance, and ethical oversight

AI risks extend beyond traditional IT risks.

They can include:

  • Bias and discrimination.
  • Privacy and data protection.
  • Cybersecurity and prompt injection.
  • Intellectual property and information misuse.
  • Inaccurate or misleading outputs.
  • Lack of transparency and explainability.
  • Unintended autonomous actions.
  • Legal, regulatory, and reputational exposure.

Directors should ensure that these risks are integrated into enterprise risk management rather than managed as isolated technology concerns.

  1. Stakeholder impact and organizational trust

AI decisions can affect employees, customers, suppliers, citizens, and communities.

For example, an AI system that automatically rejects a loan application may be efficient, but what happens if the decision is based on inaccurate or unfair data?

Governance must consider the people affected by AI, their rights, the availability of explanations, and mechanisms to challenge or correct outcomes.

  1. Competence, culture, and organizational readiness

AI governance requires more than policies. It requires people who understand the technology, its limitations, and their responsibilities.

Boards should evaluate whether the organization has sufficient AI literacy, technical expertise, risk-management capabilities, and leadership awareness.

This includes preparing project managers, business owners, procurement teams, auditors, legal professionals, and employees to work responsibly with AI.

  1. Monitoring, assurance, and continuous improvement

AI systems can change in performance as data, user behavior, models, and operating conditions evolve.

Directors should expect regular reporting on:

  • AI system performance and reliability.
  • Risk and control effectiveness.
  • Significant incidents and near misses.
  • Bias and stakeholder-impact assessments.
  • Third-party AI dependencies.
  • Changes in intended use or system capability.
  • Remediation progress and outstanding risks.

The objective is not to eliminate every possible AI risk. It is to ensure that risks are understood, governed, monitored, and addressed in a way that reflects organizational responsibilities.

  1. A practical example: The board approves an Agentic AI initiative

Imagine a global organization deploying an AI agent to support procurement.

The agent can analyze supplier proposals, compare pricing, review contract terms, identify delivery risks, and recommend preferred vendors.

The next phase allows it to initiate purchase orders within predefined limits.

Productivity improves, procurement cycles become faster, and operational costs decline.

However, several governance questions emerge:

  • Can the agent access confidential supplier information?
  • Can it make commitments on behalf of the organization?
  • What spending limits apply?
  • Can it modify supplier records?
  • How does it detect conflicts of interest?
  • Who approves high-value transactions?
  • Can the agent explain why it selected a supplier?
  • What happens if the agent is manipulated through malicious instructions?
  • Who can suspend its access when something goes wrong?

A governance-led operating model would establish clear boundaries before expanding autonomy.

This is where ISO/IEC 38507 becomes especially relevant: the governing body provides direction and oversight, while management implements the operational arrangements.

The principle: An AI agent may execute an authorized action, but it must not become an unaccountable decision-maker.

  1. What ISO/IEC 38507 means for project managers and PMOs

For project managers, the standard offers an important opportunity to elevate AI governance from a compliance activity into a core component of project delivery.

A project can meet its schedule, budget, and technical specifications while still creating unacceptable organizational risks.

PMOs should therefore integrate governance into the full AI lifecycle.

Five questions every AI project manager should ask

  1. Purpose: Is the AI use case aligned with an approved business objective?
  2. Ownership: Who is accountable for the system and its decisions?
  3. Risk: Have technical, operational, ethical, legal, and stakeholder risks been assessed?
  4. Evidence: Can we demonstrate that the AI system performs as intended?
  5. Oversight: Can humans intervene, challenge, or stop the system when necessary?

These questions can be embedded into project charters, RAID logs, steering committee reports, stage gates, vendor assessments, and benefits-realization reviews.

For PMOs, this creates a repeatable governance approach rather than allowing each AI project to develop its own interpretation of accountability.

  1. A director's AI governance dashboard: Turning oversight into evidence

Boards cannot govern what they cannot see.

A practical AI governance dashboard can help directors move from high-level assurances to evidence-based oversight.

The real measure of governance is not the number of controls documented; it is the organization's ability to demonstrate that those controls work.

  1. How ISO/IEC 38507 connects with other AI governance frameworks

ISO/IEC 38507 is most useful when considered as part of a broader governance ecosystem.

 These resources are complementary, not interchangeable.

For example, ISO/IEC 38507 helps directors understand their governance role, while ISO/IEC 42001 provides requirements for an organizational AI management system. NIST AI RMF can help operational teams structure risk-management activities.

Together, they can support a governance model connecting board oversight to operational execution and continuous assurance.

  1. The leadership opportunity: From AI adoption to AI stewardship

The next stage of AI transformation will challenge traditional governance structures.

Organizations are moving from AI tools that generate recommendations to AI agents that can initiate workflows, interact with enterprise applications, and execute actions within defined permissions.

This creates a new responsibility for directors.

They must ensure that the organization's governance model evolves at the same pace as its technological capabilities.

Boards should not wait for a major AI incident before asking how AI decisions are authorized, monitored, and challenged.

Instead, they should make AI governance a recurring boardroom discussion, supported by transparent reporting, clearly assigned accountability, and independent assurance where appropriate.

For leadership teams, this means moving beyond questions such as:

  • How much productivity will AI deliver?
  • How many processes can we automate?
  • How quickly can we deploy AI?

And asking:

  • Are we using AI for the right purpose?
  • Are our controls proportionate to the risks?
  • Can we explain and defend consequential AI-assisted decisions?
  • Are our employees equipped to oversee AI responsibly?
  • Can we demonstrate that our AI systems remain aligned with organizational values and obligations?
  1. A call to action for directors, executives, and project leaders

ISO/IEC 38507 should not sit untouched in a standards library. Its value emerges when its governance guidance influences real organizational decisions.

AI leaders can start by bringing five questions to their next board or executive steering committee meeting:

  1. Do we have a complete and current view of where AI is being used across the organization, including third-party solutions?
  2. Have we defined who owns the risks and outcomes of each significant AI use case?
  3. Are we assessing the impact of AI on customers, employees, partners, and other stakeholders?
  4. Do we have evidence that AI controls are effective throughout the system lifecycle?
  5. Can we intervene quickly when an AI system behaves outside its approved purpose or authority?

If these questions cannot be answered confidently, the organization may have an AI adoption strategy—but it may not yet have a mature AI governance model.

The opportunity for directors is to make responsible AI a board-level priority, not an afterthought delegated entirely to technology teams.

The opportunity for project managers is to translate that direction into measurable controls, accountable delivery, and evidence of responsible implementation.

The opportunity for organizations is to build trust while continuing to innovate.

Closing thought

AI governance is not about asking machines to behave responsibly. It is about ensuring that organizations take responsibility for the machines they design, procure, deploy, and use.

ISO/IEC 38507 reminds organizational leaders that AI introduces new opportunities, but also new responsibilities that cannot be delegated to algorithms or technology providers.

The future of responsible AI will not be determined solely by the intelligence of the systems organizations deploy. It will be determined by the quality of the governance, judgment, and accountability behind them.

“Boards do not need to control every algorithm. They must ensure that every consequential use of AI has a clear purpose, defined authority, accountable ownership, and evidence of responsible oversight.”

 

Search

View the archives