Who Is Accountable When AI Goes Global?

Tony Oweke is a research fellow specializing in artificial intelligence governance and digital policy at the Council on Foreign Relations. Prior to joining CFR, Tony represented a bloc of 134 developing countries in negotiations on artificial intelligence at the United Nations. 

Artificial intelligence (AI) systems are increasingly being deployed across borders with no accountability to the populations they affect. Cancer detection algorithms trained on data from high-income countries, for example, continue to misdiagnose patients across the Global South, where darker skin tones and distinct disease profiles were never represented in training datasets. Across Europe, the use of AI in border and asylum systems—including for credibility assessments, identity verification, and lie detection—raises the risk that asylum seekers could be incorrectly returned to unsafe countries and exposed to persecution or other grave human rights abuses through opaque decision-making processes.

These are not just the consequences of foreign actors operating in bad faith. They are the byproduct of deploying a new technology transnationally in the absence of shared standards, guardrails, or governance mandates commensurate with AI’s global reach.

Divergent stakeholders are converging to address that reality, and the resulting contest of ideas has sparked a governance race among regional organizations, international entities, and governments. The question of AI governance is so acute that even Pope Leo deemed it necessary to pontificate, releasing an encyclical that situates AI alongside the labor crisis of the first industrial revolution and calls for AI governance centered on accountability, transparency, and meaningful participation. This contest of ideas reflects both a shared sense of urgency and an inability to achieve meaningful consensus. The result has been a confounding patchwork of standards, frameworks, resolutions, principles, international summits, regional legislation, and even a Framework Convention on AI—all of which suffer from limited scope, exclusivity, lack of an enforcement mechanism, or a combination of the three.

This is one reason why countries are racing to establish domestic AI governance mechanisms, often in distinct ways. Major differences across jurisdictions include whether AI governance initiatives are institutionalized through voluntary frameworks or binding law or through cross-cutting or sector-specific rules, and whether authority sits with existing regulators, new coordination bodies, or legislation.

For instance, Singapore has built an AI governance architecture centered around national strategy, voluntary model frameworks, testing and assurance tools, and sector-specific regulation, but without a comprehensive AI statute. India’s 2025 AI Governance Guidelines established a principle-based, non-binding model that builds upon existing laws and institutions rather than creating a single AI act, while envisioning new coordinating bodies such as the proposed AI Governance Group and AI Safety Institute.

Brazil, by contrast, has combined its national AI strategy with draft legislation adopting a risk-based approach; the bill is currently awaiting consideration by the Chamber of Deputies. The United States’ recently released National Policy Framework for Artificial Intelligence [PDF] reflects yet another model: one that seeks to institutionalize AI governance through future federal legislation rather than various state AI laws, while relying on existing regulatory bodies with subject-matter expertise rather than a centralized AI authority. An accompanying White House statement indicates the Trump administration intends to turn this framework into legislation in the “coming months,” though that seems unlikely given deep divisions within Congress and the sheer complexity of the task.

This evolution of divergent domestic regulatory regimes poses a profound challenge for companies seeking to deploy AI systems across jurisdictions. Beyond the difficulty of complying with divergent regulatory requirements, this fragmentation undermines both public- and private-sector capacity to operate according to shared rules on attribution, responsibility, and accountability when transnational harms occur—as they already have and will continue to do. Governments around the world have consequently looked to the United Nations [PDF] as a venue for deepening collective understanding of AI’s implications and enhancing the interoperability of AI governance approaches.

The United Nations, in theory, affords all countries an opportunity to shape AI governance and ensure that AI systems are deployed across borders equitably, responsibly, and in accordance with existing international law. At minimum, the United Nations is well positioned to serve as a coordinating body for brokering transboundary agreements on AI’s role in areas such as global health, warfare, humanitarian law, climate, food security, and international development. At most, the organization can be a springboard for the global diffusion of enforceable legal instruments—be that through treaty interpretation or ratification, or through informing member states’ domestic strategies and governance frameworks.

At a moment when multilateral institutions and the international order face unprecedented challenges to their legitimacy and long-term viability, the United Nations’ next steps in AI governance may prove decisive in determining whether member states continue to view the organization as capable of meeting their needs. Through the Global Digital Compact, they have given the United Nations a clear mandate: first, to establish an effective Independent International Scientific Panel on AI that can provide the evidentiary basis for national and regional policy initiatives while informing multilateral discussions on AI governance; and second, to establish the Global Dialogue on AI Governance as a platform for policy debate and ideation. Together, these two initiatives are meant to build stronger international consensus on the most critical areas for transnational AI agreements and the United Nations’ role in brokering them.

While the United Nations has made laudable progress in designing international principles on AI safety, accountability, and transparency, effectively operationalizing those principles will be a greater challenge. For most countries, AI’s societal impacts will be determined not by theoretical capabilities developed in a lab, but by how these systems become embedded within existing social and legal infrastructures.

For most governments, the point of deployment is when the importance of accountability, due process, and oversight becomes most acute—particularly when they have limited insight and influence over how these systems are developed and implemented. The current push by many countries to establish clear rules of the road for AI governance is neither irrational nor foolhardy: it is the natural response of states seeking to protect their agency, sovereignty, and dignity in the face of a transformative transnational technology developed by a few but deployed to many. The question now is whether the most obvious multilateral body for building international consensus on AI governance will be able to deliver in a meaningful way.

This work represents the views and opinions solely of the author. The Council on Foreign Relations is an independent, nonpartisan membership organization, think tank, and publisher, and takes no institutional positions on matters of policy.