Papers · Proving training · ZKPROV

ZKPROV

ZKPROV: A Zero-Knowledge Approach to Dataset Provenance for Large Language Models

Venue
arXiv:2506.20915
Date
2025-06-26
Paper
https://arxiv.org/abs/2506.20915

Notes

Adjacent to zkPoT: proves which dataset a model was trained on, without proving the training computation itself. Cheaper, weaker claim. See also "Verifiable Fine-Tuning" (arXiv:2510.16830), which binds training proofs to a dataset commitment and policy.

Our reading

Citation neighbourhood

Builds on
  • none recorded
Cited by, in this corpus
  • none recorded

Edges are a proxy: paper A's text mentions B anywhere (body or bibliography). See the full graph.

Discussed in

A claim about a model, not about a computationThe prior surveys, and where we disagree with themThe hardest claim in the tableEvery relaxation, and what an adversary can still doThe systems, grouped by what they gave up