Overview
- The AI Provenance Assistant went live in late July and is publicly available as an early-stage tool developed by Santa Clara University professors Michael Santoro, Haibing Lu, and Michele Samorani.
- Phase one uses the Jeu de Paume / ERR Project dataset of roughly 40,000 records documenting works taken from Jews in German‑occupied France and Belgium between 1940 and 1944.
- Public tests have exposed technical flaws: some queries returned backend errors, others produced incorrectly formatted tables or links to raw .csv files that did not answer the user’s question.
- The team tailored existing large language models rather than building a proprietary model and is actively seeking provenance experts, archive partners, funders, and plans to publish research and form a foundation to hire lawyers and provenance specialists for restitution work.
- The project speaks to a wider challenge—historians estimate hundreds of thousands of works were looted in the Nazi era and tens of thousands remain unlocated—so developers say expanding archives and legal support could speed identification and return of missing works and could be adapted for other repatriation efforts.