Balancing the benefits and risks posed by artificial intelligence (AI), one of the most diffuse and rapidly evolving emerging technologies, is imperative when forming sound policy. This report analyzes the threats, trade linkages and mechanisms, and policy options in light of ongoing discussions regarding the prospects for applying export controls on artificial intelligence technologies and applications.
Using open source research, findings from organized dialogues, and expert interviews, the report authors identified policy options that go beyond export controls and encompass a coordinated, comprehensive, and technical approach to garnering the many benefits of artificial intelligence while mitigating its security risks. These approaches take into account both traditional nonproliferation strategies and ongoing debates concerning national security and economic competitiveness. Urgent, cross-sector action by governments and nongovernmental entities, including exporters, technology developers, academia, and civil society, is necessary to activate cooperative tools that mitigate the risks posed by AI. Lessons learned from strategic trade approaches to AI can be replicated, in certain situations, to other emerging technologies.
Risks
• The report authors organize AI-related risks according to risks to direct national security infrastructure, as well as indirect risks that stem from the development of the technology and nature of innovation;
• Risks to national security infrastructure include the potential of AI as an augmentation system for automation (decision-making and command and control), cyber capabilities, information and surveillance, and physical production;
• Indirect risks include asymmetric research and development progress, piecewise AI-relevant security measure implementation, poor or mismatched regulations over borders, and effects on norms, governance, and credibility.
AI and Strategic Trade Controls
• The report analyzes six possible pathways to control AI-related transfers: software, data, computing power and associated hardware, services and deemed exports, end-use/end-user controls, and catch-all controls;
• U.S. policy with regards to developing, integrating, and applying AI is described in the report and determined to be increasingly implemented against a backdrop of concerns regarding security and economic risks;
• The report finds that early list-based export control policy efforts, particularly those justified by economic competitiveness arguments, are likely to be ineffective in most situations and could make it more difficult to mitigate whatever risks AI presents. In light of this, the report reevaluates the conceptual basis for imposing controls on emerging technologies where their dangers and military end-uses are not yet known.
Based on these analyses and findings, the authors assess and where appropriate recommend the following policy options:
• Outreach
• Interagency coordination and information-sharing
• Enforcement and licensing of catch-all
• Investment controls
• Development and implementation of research criteria
• Development of norms
• Private sector self-policing/role of competition
• Targeting intangible transfers of technology (ITT)
• Technology tracking