# OriginBlame: Record- and Token-Level Data Provenance for AI Training Datasets

arXiv: https://arxiv.org/abs/2607.13037 · HTML: https://mlsanity.com/paper/2607.13037/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Machine unlearning, one of 12 directions in LLM safety and security.
- Builds on 10 of its 28 references that are in the map.
- First posted on arXiv 19 May 2026.

## Placed in

- LLM safety and security / [Machine unlearning](https://mlsanity.com/safety/machine-unlearning/)

## Builds on (most cited first, 10 of 10)

- [StarCoder: may the source be with you!](https://mlsanity.com/paper/2305.06161/) (2023)
- [TOFU: A Task of Fictitious Unlearning for LLMs](https://mlsanity.com/paper/2401.06121/) (2024)
- [The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning](https://mlsanity.com/paper/2403.03218/) (2024)
- [The Stack: 3 TB of permissively licensed source code](https://mlsanity.com/paper/2211.15533/) (2022)
- [MUSE: Machine Unlearning Six-Way Evaluation for Language Models](https://mlsanity.com/paper/2407.06460/) (2024)
- [Training Data Influence Analysis and Estimation: A Survey](https://mlsanity.com/paper/2212.04612/) (2022)
- [Fast Exact Unlearning for In-Context Learning Data for LLMs](https://mlsanity.com/paper/2402.00751/) (2024)
- [Not All Data Are Unlearned Equally](https://mlsanity.com/paper/2504.05058/) (2025)
- [Design and Evaluation of IPFS: A Storage Layer for the Decentralized Web](https://mlsanity.com/paper/2208.05877/) (2022)
- [Fine-Grained Traceability for Transparent ML Pipelines](https://mlsanity.com/paper/2601.14971/) (2026)

## Abstract (arXiv)

When a data contributor requests removal, model trainers face a practical gap: unlearning algorithms require a forget set, yet no tool can locate which training records belong to a given author. Existing provenance systems operate at file or dataset level, forcing catastrophic over-deletion. We present ob, a record- and token-level data provenance system that propagates author identity through data processing pipelines and resolves revocation requests into precise forget sets via deterministic queries. Evaluation on 219,555 Wikipedia pages demonstrates that record-level provenance eliminates dataset-level over-deletion (from 101x to 1.3x), while integration adds 1.3-4.0% throughput overhead (HuggingFace) and 2.1-19.0% (Datatrove) on wiki data. On a 1.7B model, provenance-based forget sets consistently reduce the collateral damage of machine unlearning (retain perplexity) relative to same-size random baselines across all evaluated authors, with membership-inference tests indicating they select genuinely memorized content.