AI Policy

In the context of artificial intelligence systems such as large language models (LLMs), we have established a policy to be followed internally and by anyone who wishes to contribute to ToS;DR.

Permitted use of AI

You may use AI tools for coding and for creating pull requests or patches across all of our codebases. Any such use must be disclosed in your pull request, without exception.

This requirement is not intended to shame or penalize contributors. Rather, it allows us to dedicate appropriate time to thoroughly reviewing and testing the changes, ensuring they behave as intended, are reliable, and meet our code quality standards.

You do not need to disclose AI usage when relying solely on code completion tools (such as GitHub Copilot).

Rule of thumb: if more than 20% of the work was generated by AI, it must be attributed.

AI and ML inside ToS;DR

Reviewing and interpreting terms of service and privacy policies is a sensitive task that requires careful human judgment and expertise. While we embrace careful and deliberate use of AI tools to help with the tedious parts of analysis, we take seriously their unreliability and won't jeopardize our user's trust through over-reliance. Therefore we will not:

1. Publish official ToS;DR grades and analysis points that have not gone through human review

2. Use generative LLMs to directly summarize privacy policies

3. Use LLMs or machine learning as part of our analyses without publishing technical details, and open-sourcing the code

DocBot — our ML analysis tool

Internally we have developed and deployed a machine learning tool, called DocBot, to help us with the first step of analyzing privacy documents -- highlighting cases that a user might want to know about a particular service.

All outputs generated by DocBot are always reviewed by humans before being included in the grading formula for official ToS;DR grades. Reviews are carried out by a small, trusted group of contributors with proven expertise in evaluating terms of service and privacy policies. This human-in-the-loop process ensures accuracy, consistency, and reliability, while maintaining the high standards our users expect from ToS;DR.

For technical details on DocBot, you can read a blog post here from its developer, Evan Radkoff