Governance as a Driver of Innovation in Artificial Intelligence

A global study conducted by KPMG in 2025, entitled Trust, Attitudes and Use of Artificial Intelligence, indicates that more than half (54%) of the 48,000 people interviewed across 47 countries are cautious about trusting AI systems. Seven out of ten respondents consider regulation of this technology necessary.*

Markets do not operate efficiently simply because institutions possess technical capabilities, but because their participants trust that they operate within predictable and verifiable boundaries. Investors trust financial reports, consumers expect systems not to expose them to arbitrary harm, and regulators need to believe that the entities they supervise will manage innovation in accordance with legal and institutional expectations.

The digital economy has always depended on this trust. With artificial intelligence, however, an additional requirement emerges: beyond trusting institutions, participants must trust the systems themselves and the ability of organizations to supervise them responsibly. When algorithms begin to shape the informational foundations on which organizations make decisions, the governance of these systems becomes inseparable from the governance of the institution itself.

“Trust in AI systems is inherently more fragile than the trust placed in previous forms of enterprise software. […] An algorithmic decision perceived as biased, inexplicable, or lacking accountability mechanisms can call into question not only the system’s functionality but also the legitimacy of the institution itself that uses it,” warns André Rizzo in his recent book, Reflections on AI Governance and Compliance.

Failures of this nature can have immediate repercussions for institutional reputation, as they resonate with society’s concerns about automation, inequality, and institutional power. Controversies involving algorithmic bias in hiring processes, facial recognition errors in public security, and opaque recommendation systems that influence information flows demonstrate how technical issues can turn into crises of legitimacy. In such cases, the controversy extends beyond system architecture: even a model that meets its internal metrics may lose public trust if affected individuals cannot understand how it works or challenge its outcomes.

Artificial intelligence is often presented as an objective technology capable of making decisions based exclusively on data. At first glance, algorithms appear to offer consistency in situations where human judgment may be influenced by emotions, habits, or unconscious preferences. This perception, however, overlooks the fact that AI is developed by people, trained on data produced by institutions, and implemented in environments shaped by legal, social, economic, and organizational structures.

If these data incorporate historical inequalities, incomplete representations, or biases present in previous decisions, algorithmic systems may reproduce — and even amplify — such patterns. Algorithms do not understand concepts such as equity, justice, or social responsibility. Their objective functions are generally formulated to optimize predictive capability, efficiency, or operational performance, rather than to assess whether a statistically successful pattern should be reproduced. Neutrality, therefore, is not an inherent property of technology, but a governance issue. Fairness does not emerge naturally from computational sophistication: it requires continuous evaluation, testing, and institutional oversight.

Responsible organizations establish processes to examine training data, monitor model outcomes, assess their distributive effects, and identify unintended disparities in real-world decision-making contexts. This oversight may include interdisciplinary teams, methodologies for testing fairness, continuous performance evaluation, and mechanisms for escalating and investigating questionable outcomes.

None of these mechanisms can guarantee perfectly neutral outcomes in every context. They can, however, reduce the risk of algorithmic systems silently reproducing patterns that are incompatible with institutional values or regulatory expectations. The absence of governance does not eliminate complexity: it merely postpones addressing it until a failure or public scrutiny imposes its consequences.

Governing AI, therefore, does not mean constraining innovation, but creating the conditions to sustain it. By making systems more understandable, open to challenge, and subject to accountability mechanisms, institutions strengthen the trust necessary to expand adoption. It is this trust that will enable artificial intelligence to advance with legitimacy and sustain new cycles of innovation.