Only
13 per cent of companies publicly claim adherence to a formal AI governance
framework, according to a 2025 global analysis of nearly 3,000 organisations
published jointly by UNESCO and the Thomson Reuters Foundation. That figure,
drawn from the largest corporate responsible AI dataset ever assembled,
captures the defining tension of the current moment: AI is being embedded in
products, services, and operations faster than the governance structures
needed to manage it are being built.

The same dataset, which examined 2,972 companies across 11 sectors
and multiple regions using publicly available disclosures, found that 43.7
per cent of organisations publicly communicate having an AI strategy or
guidelines. The gap between declaring intent and implementing governance is
where risk accumulates, and where the regulatory and reputational
consequences of getting AI wrong are now beginning to materialise in ways
that are no longer theoretical for any large organisation.

The Business Case for Ethics Has Changed

The argument for investing in AI ethics has shifted in character.
Where once it was framed primarily as a cost of compliance or a reputational
safeguard against worst-case scenarios, research
from the IBM Institute for Business Value offers a different
picture. In a survey of 915 global executives, IBM found that organisations
investing more heavily in AI ethics consistently achieved higher operating
profit, stronger return on investment, and measurable competitive advantage
from their AI deployments. The executives who treated ethics as
infrastructure rather than overhead outperformed those who did not, and the
performance differential was consistent across sectors.

This finding reframes the conversation in a way that is
commercially useful. The question is no longer whether to invest in ethical
AI governance but how to operationalise principles that often remain abstract
inside organisations. As our earlier coverage of AI
bias guardrails and their role in building fairer systems
illustrated, the practical work of embedding fairness and accountability into
AI systems requires specific technical and organisational measures, not
simply a public commitment to good values. Commitment and capability are
different things, and the gap between them is where reputational and legal
risk accumulates.

Where the International Framework Points

UNESCO’s
Recommendation on the Ethics of AI, adopted in November 2021 and
applicable across all 194 UNESCO member states, provides the most widely
recognised international framework for AI governance currently in existence.
Its core principles, centred on human rights, transparency, accountability,
and diversity, have become reference points for companies seeking a
governance architecture that works across multiple jurisdictions without
requiring a separate compliance programme in each.

The practical requirements the Recommendation implies are
substantive. It calls on private sector organisations to develop due
diligence and supervision mechanisms capable of identifying, preventing, and
mitigating risks throughout the AI system lifecycle, not only at the point of
initial deployment. It frames governance as a strategic responsibility that
companies in the AI value chain are expected to take on as a matter of
organisational design, not solely as a response to external pressure. For the
87 per cent of companies that have not yet publicly claimed adherence to any
formal governance framework, that expectation represents a significant gap
between what the international consensus requires and what is demonstrably in
place.

What Responsible AI Looks Like in Practice

The debates around AI ethics often operate at a level of
abstraction that makes them feel distant from operational decisions. The
harder question is what transparency, fairness, and accountability actually
require when they reach the product team, the compliance function, and the
board. Should a hiring model optimised for speed be deployed if it produces
statistically unequal outcomes across demographic groups? Should a
customer-facing system generate responses that are legally accurate but emotionally
misleading? These are not theoretical dilemmas but operational ones that land
on product managers and risk officers every day, and they require structures
for decision-making rather than principles stated on a
website.

The case of AI in legal and judicial contexts has made these
tensions visible in a pointed way. As our investigation into AI
decision-making in legal settings found, the demand for
explainability becomes most acute precisely when AI outputs affect rights and
liberties. In commercial settings the same demand exists, mediated through
customer trust and regulatory expectation rather than courtroom procedure,
but the underlying requirement is identical. A system that cannot explain why
it made a decision is a system that cannot be audited, challenged, or
corrected. That inability is not a technical limitation to be tolerated; in
an increasingly regulated environment, it is a compliance failure waiting to
be triggered.

Regulation Is Closing the Gap

The convergence between voluntary ethics frameworks and
enforceable regulation is accelerating at a pace that was not anticipated
even two years ago. The EU AI Act has moved from aspiration to enforcement
infrastructure, with prohibitions on unacceptable AI practices having been in
effect since February 2025 and general-purpose AI model obligations in force
since August 2025. The next wave of requirements, covering high-risk AI
systems in employment, credit decisions, education, and healthcare, is
scheduled for August 2026 and will apply to a far broader set of
organisations than the current GPAI obligations.

For companies operating across multiple jurisdictions, the
compliance landscape is not a single framework to navigate but a patchwork of
requirements that interact in ways that penalise organisations unprepared for
their specificity. Understanding the
shifting regulatory landscape that now governs AI deployment is
becoming a core business competency, not a periodic exercise reserved for
legal teams responding to new rules after they take effect.

What emerges from this landscape is a clearer picture of what
responsible AI requires at the business level: governance embedded in design
and deployment rather than applied retrospectively; documentation and audit
trails that can withstand regulatory scrutiny when it arrives; and leadership
that treats ethics not as a friction on innovation but as a prerequisite for
growth that can be sustained. The organisations building these capabilities
now are doing so not out of altruism but because the competitive, regulatory,
and reputational consequences of not doing so are no longer containable. The
business case for AI ethics is no longer a projection based on anticipated
regulation. It is a pattern already visible in the performance of
organisations that got there first.

The practical gap between ethics in principle and ethics in
practice is sometimes described as the “operationalisation
problem,” and it is where most internal governance efforts struggle.
Declaring commitment to human rights, transparency, and accountability in a
corporate policy document is a different undertaking from building the
processes that make those commitments real when a product manager is deciding
whether to ship a feature on a deadline. What is needed is not more
principles but more decision frameworks: structured processes for escalating
ethical trade-offs, clear accountability for the outcomes of AI-driven
decisions, and systematic review mechanisms that examine whether stated
values are reflected in actual system behaviour. Organisations that have made
this transition treat ethics review as part of product development rather
than a post-launch audit, and their compliance record under emerging
regulation reflects that structural difference.

The pattern is clear: governance built early costs less and
delivers more than governance retrofitted after problems
emerge.

About the Author

Stuart Kerr is the Technology Correspondent for LiveAIWire. He
writes about artificial intelligence, emerging technology, and the forces
reshaping work, business, and society.