Abstract
The global AI governance landscape of mid-2025 is characterized by regulatory fragmentation: the EU AI Act is in phased enforcement, the US has a patchwork of sector-specific guidance and executive orders without binding horizontal legislation, China operates under its Generative AI Interim Measures, and a growing number of countries are enacting their own frameworks with varying risk thresholds and compliance requirements. AI systems deployed across jurisdictions must navigate obligations that are sometimes contradictory - what one regime requires for transparency, another prohibits for data sovereignty reasons. This article examines the most acute jurisdictional conflicts, the mechanisms that have been proposed to manage them, and the governance gaps that remain structurally unresolved.
The Layered Conflict Problem
Jurisdictional conflict in AI governance operates at multiple layers simultaneously. At the data layer, GDPR restricts transfers of personal data to jurisdictions without adequate protection, while US national security law can compel US-headquartered companies to disclose data held anywhere in the world under CLOUD Act authority. An AI system that processes EU user data on behalf of a US cloud provider sits at the intersection of these regimes with no clean resolution.
At the model layer, the EU AI Act’s conformity assessments and transparency obligations apply to AI systems made available in the EU market regardless of where the model was trained. A system built in the US under looser requirements may need to be modified, documented, or restricted for EU deployment - but the same model version may be in use in markets that have no equivalent requirements. Version fragmentation creates compliance costs and can incentivize regulatory arbitrage, where capabilities that are restricted in high-regulation markets are simply deployed elsewhere.
At the use-case layer, applications that are permitted or lightly regulated in one jurisdiction are prohibited in another. Real-time biometric identification in public spaces is banned for law enforcement in the EU under the AI Act’s prohibited practices, while it remains operational in several US cities and is actively promoted by governments in parts of Southeast Asia and the Gulf region.
Proposed Coordination Mechanisms
Several multilateral coordination efforts are active, with limited progress. The OECD AI Policy Observatory has produced the most widely referenced framework of AI principles, endorsed by 46 countries, but the principles are non-binding and national implementations diverge substantially. The G7 Hiroshima AI Process produced a voluntary code of conduct for advanced AI developers in 2023; uptake among non-G7 developers has been limited. The Global Partnership on AI (GPAI) has working groups on responsible AI and data governance but no enforcement mechanism.
Bilateral mutual recognition agreements for AI conformity assessments - analogous to those that exist for product safety standards - have been discussed between the EU and the US but not formalized. The EU-US Trade and Technology Council, which was active through 2024, identified AI governance alignment as a priority area without producing binding outcomes. The fundamental obstacle is that the EU’s risk-based mandatory compliance model and the US’s sectoral voluntary-standards model reflect different underlying regulatory philosophies, not merely different technical specifications.
The Accountability Vacuum for Global Deployments
The most practically consequential gap is accountability for harms that occur in jurisdictions with weak or absent AI governance. A model developed by a US or EU company, deployed by a local intermediary in a country with no AI-specific regulation, used in a consequential decision that harms an individual - who bears responsibility, and through what mechanism can that individual seek redress? Existing international human rights frameworks provide normative guidance but no enforcement pathway. The UN High-Level Advisory Body on AI, reporting in 2024, identified this gap explicitly but its recommended governance architecture - an intergovernmental panel with review functions - has not advanced to institutional form. Until it does, global AI deployment operates with a structural accountability floor that varies by the regulatory ambition of whichever jurisdiction the developer happens to be incorporated in.