AI Governance Has a Human Problem: Rules Can Be Safe, Fair and Still Fail Society

Artificial intelligence governance is advancing quickly. Regulators and international institutions now speak a common language of transparency, accountability, privacy, fairness, safety, human oversight and risk management. However, a new review argues that these safeguards, while essential, are not enough for a technology reshaping how people learn, work, decide, trust and understand themselves.

The study, "Artificial Intelligence Governance and the Future of Human Societies: Dignity, Meaning and Civilizational Responsibility," published in the journal Societies, was authored by Carlos Alberto Echeverría Mayorga, Marta Irene Flores Polanco and José Miguel Esperanza Amaya of the Universidad Salvadoreña Alberto Masferrer and the Universidad Politécnica de El Salvador. It asks whether current AI governance protects not only users and data subjects, but also human agency, dignity, moral formation, social trust and the future of democratic societies.

Stronger Rules, a Thinner View of Humanity

The study finds that AI ethics has achieved substantial maturity around principles vital for identifying harm and assigning responsibility. The researchers compare six influential frameworks: UNESCO's Recommendation on the Ethics of Artificial Intelligence, the OECD AI Principles, the US National Institute of Standards and Technology's AI Risk Management Framework, the European Union AI Act, the Rome Call for AI Ethics and the G7 Hiroshima Process.

Together, these frameworks address discrimination, privacy violations, unsafe systems, opaque decisions and weak accountability. The EU AI Act provides a binding risk-based legal structure, while NIST offers a practical model for identifying and managing technical risks. UNESCO places dignity, inclusion and sustainability at the centre of AI ethics, and the OECD connects trustworthy innovation with democratic values and inclusive growth.

The review nevertheless identifies a shared limitation. Most frameworks describe people primarily as users, consumers, stakeholders, rights-holders or data subjects. Those categories are necessary for regulation, but they do not capture the whole human person: embodied, vulnerable, relational, morally responsible and shaped by culture, work and community.

The gap matters as AI becomes an infrastructure for education, employment, health care, public services, communication and political debate. A system may protect personal data and explain its output yet still encourage people to defer to automated judgment. It may avoid formal discrimination while narrowing human choice or making institutions dependent on external platforms.

The study calls for a richer understanding of human-centred AI. Governance should protect not only individual rights but also the conditions that allow people to exercise judgment, responsibility and meaningful autonomy.

AI Changes Human Agency Before It Breaks the Law

The authors conducted a structured Scopus-based review covering publications from 2019 to 2026. Their search produced 12,597 initial records. After duplicate removal, filtering and several stages of screening, 48 peer-reviewed studies were included in the final synthesis.

The largest group, comprising 18 studies, focused on anthropological issues such as dignity, autonomy, personhood, vulnerability, work and human flourishing. Eleven examined spiritual or theological questions, ten dealt mainly with AI ethics and governance, and nine considered civilizational themes including democracy, war, transhumanism and existential risk.

The distribution shows that concerns about dignity, meaning and social futures are already present in scholarship. The problem is fragmentation: they remain divided across disciplines and are rarely translated into shared governance criteria.

The anthropological literature shows how AI can reshape autonomy without openly coercing anyone. Recommendation systems determine which choices become visible. Predictive tools influence which people receive attention, credit, employment or public support. Automated systems can also distribute responsibility so widely that no actor feels fully accountable for the outcome.

"Human oversight" may therefore be more apparent than real. A doctor, teacher, civil servant or worker may technically retain final authority while lacking the time, information or institutional backing needed to challenge an automated recommendation.

The same concern applies to employment. AI can remove repetitive tasks and improve access to information, but it can also deskill professions, intensify surveillance and redefine good work as measurable output. A conventional audit may ask whether a system is accurate and non-discriminatory. A deeper assessment would ask whether it preserves professional judgment, dignity and responsibility. This is particularly important in health care, education and public administration, where decisions involve vulnerability, care and consequences that cannot be reduced to efficiency scores.

Meaning and Democracy Are Becoming AI Governance Issues

The study treats spiritual and civilizational questions, which are often marginal to mainstream technology policy. The spiritual dimension is not limited to religious belief. It includes meaning, interior life, attention, moral formation and discernment. AI is entering religious education, emotional support, pastoral communication and moral deliberation. These applications may expand access to information, but they can also simulate intimacy, commercialize personal reflection or encourage users to outsource difficult ethical choices.

Education presents a similar dilemma. Schools and universities are focused on plagiarism, privacy and assessment. The review suggests that they must also ask how AI affects intellectual independence, patience, critical thinking and the formation of judgment. A student may produce an acceptable answer more efficiently while engaging less deeply with the reasoning required to understand it.

At the civilizational level, the study links AI to disinformation, cognitive warfare, democratic distrust, autonomous weapons and concentrated technological power. These threats are usually framed as security or misuse problems. The authors argue that governance must also examine the kind of society that large-scale AI adoption normalizes.

Synthetic media and automated propaganda do not merely spread falsehoods. They can weaken confidence in shared truth. Similarly, an individual AI product may be safe while its widespread use increases institutional dependence on a small number of corporations controlling models, data and computing infrastructure.

The concern is especially relevant to Latin America and other parts of the Global South. Countries adopting systems developed abroad may have limited influence over the cultural assumptions encoded in them. Dependence on foreign cloud infrastructure and proprietary models can also affect data sovereignty, domestic innovation and public-sector capacity.

From Technical Compliance to Civilizational Responsibility

To bring these concerns together, the study proposes the Anthropological, Spiritual and Civilizational Framework, or ASC Framework. Its anthropological dimension asks what conception of the human person an AI system assumes or promotes. Does it preserve dignity, agency, embodied experience and meaningful work, or treat people mainly as predictable profiles and functional capacities?

The spiritual dimension asks whether AI supports reflection, attention and moral formation or replaces difficult human processes with automated guidance and simulated connection. The civilizational dimension asks what collective future the technology helps produce. Does it strengthen truth, democracy, peace, justice and cultural plurality, or intensify manipulation, dependency and concentrated power?

The framework is designed to complement, not replace, established governance tools. Transparency, fairness, safety and privacy remain the baseline. ASC adds questions about the social purposes those protections should serve.

A university assessing a generative AI platform could examine privacy and academic integrity alongside student agency, intellectual formation and dependence on an external provider. A government evaluating an automated public-service system could assess accuracy and bias while also considering dignity, appeal rights, public trust and the distribution of decision-making power.

The framework remains conceptual. It has not been validated through case studies, expert consultation or measurable indicators. The review also relied on Scopus alone, potentially missing relevant legal, engineering, medical and philosophical research. Books, reports and conference proceedings were largely outside the formal corpus, and no formal inter-rater reliability coefficient was calculated.

Future research should test ASC through Delphi studies, institutional applications and cross-cultural comparisons. Researchers will also need practical indicators for agency, dependency, moral formation, cultural plurality and democratic trust without reducing complex human qualities to misleading scores.