The Terms and Conditions of Generative AI

Authors:
John M. Newman, University of Memphis
Andres Sawicki, University of Miami

Abstract:

Generative AI raises pressing questions for myriad doctrinal areas. Policymakers and scholars have responded rapidly, devoting substantial attention to these difficult legal and policy questions. But, as a practical matter, firms offering generative AI tools can unilaterally resolve or sidestep many issues using online terms and conditions. These terms, though seldom read or understood by users, can radically alter rights, shift legal risks and costs, and eliminate entire causes of action. Yet despite their fundamental implications for law and policy, little is known about the content of generative AI terms and conditions.

This Article offers the first empirical examination, at scale, of the digital strings attached to widely available generative AI tools. Our study reveals some surprisingly user-friendly findings, as well as novel variations on familiar drafting tools like mandatory-arbitration provisions. On the whole, however, the results are troubling. Broad, royalty-free generative AI licenses threaten to erode authors’ ability to exercise copyrights in their original works. Contrary to public messaging from high-profile providers, their terms and conditions do not indemnify ordinary users against infringement risk. More broadly, an array of dispute-resolution terms limit access to adjudication, eliminate long-established causes of action, tilt adjudication in providers’ favor, and sharply limit monetary remedies for injured users.

These terms and conditions appear likely to have negative individual-, class-, and system-level impacts. Though the study’s primary contribution is empirical, the findings support immediate policy responses. We thus conclude by identifying a range of prescriptive reforms—voiding certain one-sided provisions and, under certain circumstances, imposing penalties to disincentivize their use—to ameliorate the harms currently facilitated by the terms and conditions of generative AI.