Reported by DAVID PLOTINSKY and GIOVANNA M. CINELLI

In the absence of overarching regulation in the United States, AI is currently governed by a mix of the federal government, state governments, industry itself, and the courts. These tools, however, are limited and have challenges of their own, including possible conflicts of interest in the case of industry self-governance, compliance issues that can arise from overlapping or conflicting regulations by multiple state governments, and the limitations on courts to adjudicate AI-related disputes within the confines of existing law. 

The United States does have certain existing regulatory tools that it is leveraging to address AI, in addition to developing new regulations to manage AI-associated risks. It is important for companies operating in or doing business with the United States to understand the country’s current AI regulatory landscape, which includes both the executive branch’s development of regulatory authorities and the investigative and legislative activities of US Congress.

US EXPORT CONTROLS LAWS AND REGULATIONS

AI has extensive military, defense, and intelligence capabilities, from the use of autonomous vehicles to the collection of intelligence data to the evaluation, analysis, and synthesis of huge language models and data points.

As of now, US export controls currently do not specifically “control” AI as a broad category. Instead, the different components that contribute to the development of AI are controlled in a variety of ways, the majority of which still fit in an uncontrolled or a lightly controlled category of the Export Administration Regulations (EAR99). 

These components include, but are not limited to,

• integrated circuits/semiconductors; 
• technology for designing, developing, adapting, or embedding AI functionality into products or platforms;
• equipment to manufacture the integrated circuits/semiconductors used for AI functionality; and 
• assistance deemed to be “US support” or facilitation in these areas and tangentially covered by other direct or indirect items or activities.

While AI also has several commercial applications through the proliferation of generative platforms, the technology has not yet reached a level of sufficient integrity to cede its functionalities to a standalone AI category. However, the speed and relative open development environment have raised significant concerns among the United States and other governments regarding the potential for misuse. This has led to questions of whether certain AI technologies would be more appropriately controlled under the International Traffic in Arms Regulations (ITAR), rather than the EAR99. While there are elements of AI overall, such as hardware, types of software modules, or technical data on system design and development, that could find their way onto ITAR’s US Munitions List, the ITAR does not currently include specific AI defined items (whether products, technology or software) that are subject to export control requirements. 

One of the key challenges facing any potential regulatory scheme is definitional (i.e., exactly how does one define AI?). A variety of federal agencies and Congress have proposed definitions and while some common elements exist, the distinctions raise questions regarding the ability of the agencies drafting the regulations to find the definition that works best within the regulatory framework. 

For example, in 2018, the US Department of Defense issued its first AI Strategy. The definition, however, was ubiquitous, and therefore, somewhat unhelpful. Soon after, the Fiscal Year 2023 National Defense Authorization Act (FY23 NDAA) provided a more granular definition with three levels of AI—basic, reactive, and capable of “thinking,” with the third level being the most difficult to determine whether export controls can be used. The FY23 NDAA examined the functionality and potential applications of AI and some terminology within the definition, such as “acting rationally,” has caused consternation. 

The US government remains concerned with the development of AI systems that enable the military modernization of countries of concern—including weapons, intelligence, and surveillance capabilities—and that have applications in areas such as cybersecurity and robotics.

Read full report: https://www.morganlewis.com/pubs/2024/04/existing-and-proposed-federal-ai-regulation-in-the-united-states