Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Compliance Rating Scheme: A Data Provenance Framework for Generative AI Datasets

Matyas Bohacek, Ignacio Vilanova Echavarri

TL;DR

The paper tackles the lack of transparency and accountability in generative AI training data by introducing the Compliance Rating Scheme (CRS) and an open-source library, DatasetSentinel, that embeds data provenance into dataset construction and evaluation. CRS defines six criteria spanning data sourcing, licensing, opt-outs, change traceability, and provenance augmentation, translating into a letter-grade score from A to G to guide practitioners. DatasetSentinel provides per-data-point and dataset-wide assessments, compatible with major ML ecosystems and leveraging C2PA/CAI provenance standards to enable reproducible, auditable data use. The work is demonstrated through case studies and user evaluations, highlighting both the promise of improved governance and the need for broader provenance adoption and platform integration for real-world impact.

Abstract

Generative Artificial Intelligence (GAI) has experienced exponential growth in recent years, partly facilitated by the abundance of large-scale open-source datasets. These datasets are often built using unrestricted and opaque data collection practices. While most literature focuses on the development and applications of GAI models, the ethical and legal considerations surrounding the creation of these datasets are often neglected. In addition, as datasets are shared, edited, and further reproduced online, information about their origin, legitimacy, and safety often gets lost. To address this gap, we introduce the Compliance Rating Scheme (CRS), a framework designed to evaluate dataset compliance with critical transparency, accountability, and security principles. We also release an open-source Python library built around data provenance technology to implement this framework, allowing for seamless integration into existing dataset-processing and AI training pipelines. The library is simultaneously reactive and proactive, as in addition to evaluating the CRS of existing datasets, it equally informs responsible scraping and construction of new datasets.

Compliance Rating Scheme: A Data Provenance Framework for Generative AI Datasets

TL;DR

The paper tackles the lack of transparency and accountability in generative AI training data by introducing the Compliance Rating Scheme (CRS) and an open-source library, DatasetSentinel, that embeds data provenance into dataset construction and evaluation. CRS defines six criteria spanning data sourcing, licensing, opt-outs, change traceability, and provenance augmentation, translating into a letter-grade score from A to G to guide practitioners. DatasetSentinel provides per-data-point and dataset-wide assessments, compatible with major ML ecosystems and leveraging C2PA/CAI provenance standards to enable reproducible, auditable data use. The work is demonstrated through case studies and user evaluations, highlighting both the promise of improved governance and the need for broader provenance adoption and platform integration for real-world impact.

Abstract

Generative Artificial Intelligence (GAI) has experienced exponential growth in recent years, partly facilitated by the abundance of large-scale open-source datasets. These datasets are often built using unrestricted and opaque data collection practices. While most literature focuses on the development and applications of GAI models, the ethical and legal considerations surrounding the creation of these datasets are often neglected. In addition, as datasets are shared, edited, and further reproduced online, information about their origin, legitimacy, and safety often gets lost. To address this gap, we introduce the Compliance Rating Scheme (CRS), a framework designed to evaluate dataset compliance with critical transparency, accountability, and security principles. We also release an open-source Python library built around data provenance technology to implement this framework, allowing for seamless integration into existing dataset-processing and AI training pipelines. The library is simultaneously reactive and proactive, as in addition to evaluating the CRS of existing datasets, it equally informs responsible scraping and construction of new datasets.
Paper Structure (33 sections, 7 figures, 3 tables)

This paper contains 33 sections, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Generative AI
  4. Datasets
  5. Dataset Life Cycle and Stakeholders
  6. Challenge to Address
  7. Dataset Principles
  8. Data Provenance
  9. Compliance Rating Scheme
  10. Library
  11. Features
  12. Feature 1
  13. Feature 2
  14. Dataset-level Criteria
  15. Data Point-level Criteria
  16. ...and 18 more sections

Figures (7)

  • Figure 1: A schematic overview of the AI ecosystem workflow with the main stages of dataset and model development
  • Figure 2: Proposed design interface for "A" and "C" score on the CRS scale
  • Figure 3: A schematic overview of DatasetSentinel's use case within the dataset curation stage of the dataset lifecycle.
  • Figure 4: A schematic overview of CRS' use case within the dataset repository stage of the dataset lifecycle.
  • Figure 5: A fictitious "A" CRS score mock-up of a random GitHub dataset
  • ...and 2 more figures