The entry of artificial intelligence into public institutions acts as a systemic stress test for the contemporary state. It does not merely introduce new operational tools, but challenges frameworks designed to govern industrial economies characterized by long time horizons, sequential processes and relatively stable margins of predictability. AI compresses decision time, expands informational complexity and operationalizes correlations that escape traditional causal logic. In this context, institutions are not simply called upon to modernize, but to confront a technology that exposes structural limits that have long remained latent.
Institutional capacity represents the first point of friction. Public administrations founded on procedures, hierarchies and formalized responsibilities are designed to ensure continuity and impartiality. Artificial intelligence instead introduces a logic of continuous learning and real time response. According to OECD data, more than 70 percent of governments in advanced economies are experimenting with or using AI systems in areas such as welfare, taxation, healthcare and security. However, in most cases, adoption proceeds more rapidly than the ability to integrate these systems into clear frameworks of responsibility. The temporal mismatch between public decision making and algorithmic computation risks reducing the relevance of the state as an actor capable of orienting complex economic and social processes.
Traditional administrative models reveal their limits here. Modern bureaucracy, in its Weberian form, is based on abstract rules, defined competences and procedural controls. This structure has ensured predictability and legitimacy for decades. When confronted with systems that operate through probabilistic correlations and emergent patterns, procedural rigidity can turn into a constraint. In numerous cases documented by European and North American oversight authorities, automated evaluation systems are adopted as decision support without a full capacity for internal audit. The decision remains formally public, but the logic that produces it becomes opaque, generating a grey zone in which power appears distributed but is not truly controlled.
New types of vulnerability thus emerge. Artificial intelligence promises efficiency, but introduces deep dependencies on technological infrastructures, private providers and highly specialized skills. World Bank reports highlight that most public administrations lack sufficient resources to independently develop and maintain complex systems, relying on external actors for data, models and maintenance. Institutional sovereignty does not disappear dramatically, but weakens on the cognitive and operational level. Data governance, algorithmic security and the ability to contest automated decisions become critical issues, even in the absence of visible failures.
This dependence is reflected in the relationship between the state and citizenship. Decisions taken or influenced by models that are difficult to explain erode transparency, a central element of democratic legitimacy. World Economic Forum studies indicate that trust in institutions tends to decline when decision processes are perceived as technical and inaccessible, even if formally efficient. AI, when used without adequate mechanisms of explainability and control, risks amplifying this distance, transforming efficiency into a source of political alienation.
The resulting challenge requires a redefinition of the role of the state. It is not a matter of choosing between adopting or rejecting artificial intelligence, but of recognizing that the state cannot limit itself to being a final user of technologies developed elsewhere. The public function shifts toward the design of frameworks, the definition of principles and the guarantee of intelligible processes. In an economy in which power is increasingly exercised through control over informational flows and computational architectures, sovereignty takes on a cognitive dimension. Governments and institutions are called upon to oversee not only decisions, but the conditions of possibility of decisions themselves.
This tension recalls Max Weber’s reflections on the formal legal rationality of the modern state. Weber saw bureaucracy as the culmination of Western rationalization, but also intuited its risk of rigidity. Artificial intelligence extends instrumental rationality beyond the human domain, transforming decision making into a problem of continuous optimization. The risk is not a loss of efficiency, but a hyper rationalization that reduces the space for political judgment and deliberation. Public decision making tends to be evaluated for its statistical coherence rather than for its substantive legitimacy.
The experience of the private sector offers useful indications. Firms have long operated in environments characterized by high complexity and decision speed. They have learned that automation does not eliminate responsibility, but makes it more difficult to exercise. Algorithmic errors, embedded biases and technological dependencies have shown how efficiency without governance produces vulnerability. Similarly, the state is called upon to develop an institutional leadership capable of integrating technology and judgment, data and values, computation and responsibility. This is not a matter of adopting corporate models, but of recognizing the cognitive limits of systems and building adequate counterpowers.
Artificial intelligence does not mark the end of the modern state, but puts its deep structure to the test. It exposes institutions to a verification that concerns their ability to govern not only through norms and procedures, but through invisible architectures of computation that orient behaviors and expectations. In this stress test, the boundaries between efficiency and legitimacy, between decision and responsibility, are redefined, outlining a terrain on which public governance is called to rethink its role within an economic ecosystem increasingly dominated by systems that compute with growing precision and understand to a limited extent.
Global AI Observatory