AI’s Next Governance Battle Is About Power, War and Who Gets to Set the Limits

Artificial intelligence (AI) is increasingly being treated not simply as a technology to regulate, but as a source of political, economic and security power that may require enforceable limits. UN High Commissioner for Human Rights Volker Türk has called for binding rules, independent oversight and clearer safeguards, warning that sufficiently powerful AI systems could become difficult even for their developers to control.

His intervention pushes the governance debate into harder territory. The concern is not merely confined to biased algorithms, privacy failures or unsafe software. It now encompasses who controls advanced AI, how much authority private developers should exercise, whether lethal decisions should ever be delegated to machines, and whether governments can build meaningful safeguards before technological capabilities outpace regulation.

The AI debate is moving from safety checks to enforceable limits

Türk's warning reflects a growing distinction between managing technical risk and establishing political control over powerful technologies. He argued that voluntary arrangements are not enough, calling instead for binding rules, independent verification and "cast iron guarantees" around AI safety and security.

The outcomes are significant. Voluntary commitments depend heavily on companies defining their own thresholds for acceptable risk, while enforceable regulation transfers at least part of that authority to governments, regulators or independent bodies. The question becomes not only whether AI developers can make systems safer, but whether society should rely on developers themselves to decide what levels of capability and risk are acceptable.

Türk raised extreme scenarios in which AI could escape controlled testing environments or resist attempts to shut it down. Such possibilities remain part of a wider debate over advanced AI risks, but his broader argument is institutional: regulation should be designed before increasingly capable systems create problems that become substantially harder to contain.

This creates a difficult policy challenge. Governments must determine how to regulate technologies whose future capabilities remain uncertain while avoiding rules built around speculation rather than demonstrable risk. However, waiting for serious failures before creating enforceable safeguards carries its own dangers, particularly where systems may affect critical infrastructure, public decision-making or national security.

The deeper concern is the concentration of technological power

His comments also highlight an issue that receives less attention than AI performance: control. The warning that "a handful of men" hold extraordinary influence over AI points to the concentration of computing capacity, data, technological expertise and advanced models among a relatively small group of companies and individuals.

The concentration changes the character of the governance problem. When technologies capable of influencing labour markets, information systems and public institutions are developed primarily inside private companies, decisions with potentially wide social consequences can be made outside traditional systems of democratic accountability.

Data is vital to that imbalance. Advanced AI depends on enormous information flows, while companies controlling models and infrastructure can acquire capabilities that governments, regulators and smaller competitors may struggle to match. The result is a widening gap between those designing AI systems and the institutions expected to supervise them.

Autonomous weapons expose the limits of leaving humans outside the loop

Nowhere is the question of human control more consequential than warfare. Türk called for the urgent prohibition of weapons capable of taking lives without human involvement, arguing that lethal decisions should not be transferred entirely to machines.

He referred to reports that fully autonomous drones deployed by Russia had killed three Ukrainians, while Ukraine had also reportedly tested autonomous weapons. Those reports underline a wider concern: once artificial intelligence moves from assisting military decision-making to independently selecting or attacking targets, accountability becomes significantly more difficult to establish.

The issue extends beyond whether an autonomous system performs accurately. Decisions involving lethal force carry legal, ethical and humanitarian responsibilities that traditionally rest with identifiable human actors. Greater machine autonomy can make responsibility less clear when mistakes occur, civilians are harmed or targeting decisions are challenged.

Türk's warning comes at a particularly volatile moment. He said more than 60 conflicts involving at least one country are occurring worldwide, alongside continuing civilian casualties in Ukraine and widening tensions elsewhere. The proliferation of increasingly autonomous military systems into such an environment could create risks that conventional AI safety regimes were never designed to address.

Global governance will matter only if countries can agree on where the boundaries lie

Türk's demand for cooperation between countries hosting AI technologies and those participating in their supply chains acknowledges a basic reality: advanced artificial intelligence cannot be governed effectively within national borders alone. Models, computing infrastructure, components, investment and data increasingly operate across interconnected jurisdictions.

However, international cooperation may also be the hardest part of the equation. Governments have different technological capabilities, security priorities and commercial interests. Countries competing for leadership in AI may resist restrictions they believe could weaken their strategic position, while states without major AI industries may demand stronger protections against systems developed elsewhere.

Independent verification presents another challenge. Effective supervision may require regulators or external bodies to examine how powerful systems behave, how they are tested and what safeguards developers have implemented. That raises difficult questions about proprietary technology, national security and how much access companies should be required to provide.

The same tension applies to autonomous weapons. Calls for prohibition can establish a political principle, but effectiveness ultimately depends on definitions, verification and international compliance. As autonomous capabilities become more sophisticated, distinguishing systems that assist human operators from systems that effectively make lethal decisions themselves may become increasingly important.

What emerges from Türk's intervention is a broader shift in how AI governance is being framed. The critical question is becoming less about whether AI should be regulated and more about which forms of technological power societies are prepared to tolerate without binding oversight.