Blab Tech
Generative-AI Terms and Conditions Shift Legal Risk onto U.S. Users, Study Finds
In a fresh working paper, law scholars John Newman of the University of Memphis School of Law and Andres Sawicki of the Miami School of Law expose a pattern in generative‑AI contracts that pushes significant legal responsibility onto the very people who use the services.
The researchers examined 100 terms‑and‑conditions documents from major AI platforms—including OpenAI, Anthropic, and other providers of large‑language‑model chatbots and text‑to‑image tools. Their analysis shows that 84.9 % of the contracts grant the provider a non‑exclusive license to use user inputs.
In many cases, that license is far more expansive than a simple use right. Sixty‑three percent of the contracts allow the provider to sublicense the material, 30.1 % permit use for any business purpose, and 13.7 % impose no limits on how the provider can use the content. The result is that when a user uploads a document or image, the AI company can copy, modify, and redistribute that material without paying the user or seeking additional permission. The licenses are typically irrevocable and perpetual, giving the provider a permanent right to exploit the content.
The paper also turns its eye to how AI‑generated outputs are owned. Two‑thirds of the contracts that address output ownership name the user as the owner, but the provider retains the right to use the output “however they please.” The authors warn that this arrangement can leave users with little control over how their AI‑created content is used.
Copyright liability is another point of concern. Ninety‑six percent of the contracts allow the provider to modify the terms without the user’s consent, and 88 % permit the provider to change the terms while the user cannot. Forty‑two percent of the contracts do not require the provider to give meaningful notice of changes, making it difficult for users to know when their rights have been altered.
Liability for copyright infringement also falls heavily on users. Sixty‑nine point eight percent of the T&C require users to pay if the provider’s service leads to a copyright claim against the provider. Because generative‑AI models can produce content that closely resembles existing works, users may be exposed to legal costs even when they attempt to avoid infringement.
The authors describe these unilateral provisions as a “copyright crumple zone” for consumers and argue that they undermine the ability of courts to develop balanced precedent. They call for a rapid policy response that would prohibit T&C that limit user power, shift liability to the AI company, and require clear notice of any changes.
Beyond contractual analysis, the study urges an open, deliberative process to examine how generative‑AI tools affect copyright, free speech, privacy, and antitrust law. The authors suggest that the legal community and lawmakers should consider these issues before the industry’s contractual practices become entrenched.
For now, U.S. consumers using generative‑AI services do so under contracts that often transfer significant rights and risks to the provider. The paper’s findings highlight a gap between the rapid deployment of AI tools and the development of a regulatory framework that protects users. The authors’ recommendations remain unimplemented, and it is unclear whether lawmakers will act before the industry expands further. The study underscores the need for clear rules that balance innovation with consumer protection.
The researchers examined 100 terms‑and‑conditions documents from major AI platforms—including OpenAI, Anthropic, and other providers of large‑language‑model chatbots and text‑to‑image tools. Their analysis shows that 84.9 % of the contracts grant the provider a non‑exclusive license to use user inputs.
In many cases, that license is far more expansive than a simple use right. Sixty‑three percent of the contracts allow the provider to sublicense the material, 30.1 % permit use for any business purpose, and 13.7 % impose no limits on how the provider can use the content. The result is that when a user uploads a document or image, the AI company can copy, modify, and redistribute that material without paying the user or seeking additional permission. The licenses are typically irrevocable and perpetual, giving the provider a permanent right to exploit the content.
The paper also turns its eye to how AI‑generated outputs are owned. Two‑thirds of the contracts that address output ownership name the user as the owner, but the provider retains the right to use the output “however they please.” The authors warn that this arrangement can leave users with little control over how their AI‑created content is used.
Copyright liability is another point of concern. Ninety‑six percent of the contracts allow the provider to modify the terms without the user’s consent, and 88 % permit the provider to change the terms while the user cannot. Forty‑two percent of the contracts do not require the provider to give meaningful notice of changes, making it difficult for users to know when their rights have been altered.
Liability for copyright infringement also falls heavily on users. Sixty‑nine point eight percent of the T&C require users to pay if the provider’s service leads to a copyright claim against the provider. Because generative‑AI models can produce content that closely resembles existing works, users may be exposed to legal costs even when they attempt to avoid infringement.
The authors describe these unilateral provisions as a “copyright crumple zone” for consumers and argue that they undermine the ability of courts to develop balanced precedent. They call for a rapid policy response that would prohibit T&C that limit user power, shift liability to the AI company, and require clear notice of any changes.
Beyond contractual analysis, the study urges an open, deliberative process to examine how generative‑AI tools affect copyright, free speech, privacy, and antitrust law. The authors suggest that the legal community and lawmakers should consider these issues before the industry’s contractual practices become entrenched.
For now, U.S. consumers using generative‑AI services do so under contracts that often transfer significant rights and risks to the provider. The paper’s findings highlight a gap between the rapid deployment of AI tools and the development of a regulatory framework that protects users. The authors’ recommendations remain unimplemented, and it is unclear whether lawmakers will act before the industry expands further. The study underscores the need for clear rules that balance innovation with consumer protection.
