On October 30, 2023, the White House published the "Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence" ("Order"). The volume, scope, and goals of the Order are sweeping and far-reaching, tapping the Department of Homeland Security, Department of Energy, Department of Defense, Department of State, Department of Justice, Department of Transportation, and Department of Education, among others, to assess and develop a framework to govern AI development, use, and regulation across as many industries and fields as possible.
The Order includes a number of sections that address eight guiding principles and priorities, including those relating to privacy, civil rights, and safety and security, among other topics.
These sections establish deadlines and goals for various departments for creating rules, regulation, programs, and even information collection requirements for managing AI. The targets for creating rules, regulations, and information collection requirements span commercial businesses, government agencies, health care systems, the justice system, and infrastructure. The goal of the Order is to be as comprehensive as possible, and to prepare for any developments and applications of AI.
Although the shortest timeframe is listed at 30 days from the Order, the majority of the directives provide 180 and 270 days from the Order as initial deadlines. The expectation is that wide sweeping requirements are on the horizon and may be imposed on both private and public sector use and development of AI.
What Steps Should You Take?
For clients managing AI use and development, it is imperative that you continue to monitor the results of the Order and take steps to comply.
Some proactive steps that clients can take in their AI implementation include careful management and documentation of model generation, tailoring models to provide constraints on use, and managing training data carefully, including knowing the attribution of training data. Some developers may consider synthesizing their own data for use in training and model development to avoid potential pitfalls.
Presumably, the strictest requirements will be imposed on any "dual-use foundation model." This is defined as a model that is trained on broad data, includes self-supervision, includes billions of parameters, is widely applicable, and exhibits a serious risk to security, public health or safety, or any such combination. The utmost of care must be exercised if your AI development may fall into this class.
IP counsel will eagerly await the guidance proposed for the USPTO addressing inventorship and the use of AI (120 days from Order), guidance on the intersection of AI and IP (270 days), and a forthcoming study on copyright issues raised by AI (270 days or later).
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
