Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Acceptable Use Policies for Foundation Models

Kevin Klyman

Abstract

As foundation models have accumulated hundreds of millions of users, developers have begun to take steps to prevent harmful types of uses. One salient intervention that foundation model developers adopt is acceptable use policies: legally binding policies that prohibit users from using a model for specific purposes. This paper identifies acceptable use policies from 30 foundation model developers, analyzes the use restrictions they contain, and argues that acceptable use policies are an important lens for understanding the regulation of foundation models. Taken together, developers' acceptable use policies include 127 distinct use restrictions; the wide variety in the number and type of use restrictions may create fragmentation across the AI supply chain. Developers also employ acceptable use policies to prevent competitors or specific industries from making use of their models. Developers alone decide what constitutes acceptable use, and rarely provide transparency about how they enforce their policies. In practice, acceptable use policies are difficult to enforce, and scrupulous enforcement can act as a barrier to researcher access and limit beneficial uses of foundation models. Nevertheless, acceptable use policies for foundation models are an early example of self-regulation that have a significant impact on the market for foundation models and the overall AI ecosystem.

Acceptable Use Policies for Foundation Models

Abstract

As foundation models have accumulated hundreds of millions of users, developers have begun to take steps to prevent harmful types of uses. One salient intervention that foundation model developers adopt is acceptable use policies: legally binding policies that prohibit users from using a model for specific purposes. This paper identifies acceptable use policies from 30 foundation model developers, analyzes the use restrictions they contain, and argues that acceptable use policies are an important lens for understanding the regulation of foundation models. Taken together, developers' acceptable use policies include 127 distinct use restrictions; the wide variety in the number and type of use restrictions may create fragmentation across the AI supply chain. Developers also employ acceptable use policies to prevent competitors or specific industries from making use of their models. Developers alone decide what constitutes acceptable use, and rarely provide transparency about how they enforce their policies. In practice, acceptable use policies are difficult to enforce, and scrupulous enforcement can act as a barrier to researcher access and limit beneficial uses of foundation models. Nevertheless, acceptable use policies for foundation models are an early example of self-regulation that have a significant impact on the market for foundation models and the overall AI ecosystem.
Paper Structure (9 sections, 2 figures, 2 tables)

This paper contains 9 sections, 2 figures, 2 tables.

Table of Contents

  1. 1. Introduction
  2. 2. Background
  3. 3. Methodology
  4. 4. Analysis of acceptable use policies
  5. 5. Enforcement of acceptable use policies
  6. 6. Discussion
  7. 7. Areas for Future Work
  8. 8. Conclusion
  9. Acknowledgments

Figures (2)

  • Figure 1: Common prohibited content categories and number of prohibited uses per developer. Top left: the 10 most common categories of content-related prohibited uses in developers' AUPs. Top right: the next 10 most common categories of content-related prohibited uses in developers' AUPs. (See the https://github.com/kklyman/aupsforfms/tree/main for details on grouping.) Bottom left: the number of explicitly prohibited uses in closed developers' AUPs (out of 127 categories). Bottom right: the number of explicitly prohibited uses in open developers' AUPs.
  • Figure 2: Developer correlations. The correlation between prohibited use categories for pairs of developers across all 127 categories. Correlation is measured using the simple matching coefficient (i.e. agreement rate), which is the fraction of all indicators for which both developers are assigned the same value (i.e. where both are assigned 1 as both of their AUPs prohibit the category, or both are assigned 0).