BRENNEN

Industries / Media, arts and recreation / Museums

Ownership history

Tracing who owned an object before the museum (its provenance), and returning objects that were taken.

5

Milestones achieved, from 1970 to 2024.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1970

    A treaty against trafficking

    UNESCO's 1970 Convention on the illicit import, export and transfer of ownership of cultural property "supports the fight against illicit trafficking of cultural property and facilitates its return and restitution". UNESCO counts 150 states parties.

    unesco.org

    Credit

    UNESCO

  2. 1998

    Rules for Nazi-looted art

    On 3 December 1998 the 44 governments at the Washington Conference on Holocaust-Era Assets endorsed eleven non-binding principles: art confiscated by the Nazis and not returned should be identified, archives opened, findings published, and a just and fair solution sought with the pre-war owners or their heirs.

    lootedartcommission.com

    Credit

    the governments at the Washington Conference on Holocaust-Era Assets

  3. 2007

    The Getty returns 40 objects

    In August 2007 the J. Paul Getty Museum agreed to return 40 objects to Italy. This followed a raid by Italian and Swiss police in 1995 on the dealer Giacomo Medici's storerooms in the Geneva Freeport and, in November 2005, the charging in an Italian court of the Getty's former antiquities curator, Marion True; the Los Angeles Times had reported that 350 of the museum's antiquities came from dealers suspected or convicted of dealing in looted objects.

    LessonAn object bought without a documented history can be lost again, and the museum's reputation with it.

    traffickingculture.org

    Credit

    J. Paul Getty Museum; Italian Ministry of Culture

  4. 2020

    Abomey's treasures go back

    France's National Assembly voted unanimously in October 2020 for a law, adopted on 24 December 2020, to return 26 objects taken in "the 1892 pillaging of the palace of Abomey" to Benin, and a sword to Senegal, an exception to the French rule that objects in state collections cannot be given away. The 26 works were shown at a ceremony at the Musée du Quai Branly in Paris on 27 October 2021 before ownership passed to Benin. The move followed President Emmanuel Macron's 2017 speech in Ouagadougou and a report by Felwine Sarr and Bénédicte Savoy, which put the African holdings of French museums at 90,000 objects.

    artreview.comelysee.fr

    Credit

    French Parliament; Felwine Sarr; Bénédicte Savoy

  5. 2024

    Consent before display

    On 12 January 2024 revised US rules under the Native American Graves Protection and Repatriation Act (NAGPRA) took effect, requiring museums to obtain consent before exhibiting or researching human remains and cultural items covered by the law. The Field Museum in Chicago covered display cases, and the American Museum of Natural History, the Metropolitan Museum of Art, Harvard's Peabody Museum and the Cleveland Museum of Art removed objects from view or covered cases.

    LessonGalleries built without consulting the people the objects came from can be emptied overnight when consent becomes the rule.

    theartnewspaper.com

    Credit

    the revised NAGPRA rules; the museums named

Watch for

  1. Returns by decree in France

    France's framework law on returning cultural property lets the state return objects acquired between 1815 and 1972 by decree, after advice from a scientific committee and a national restitution commission, instead of passing a separate law for each return; it requires establishing whether each acquisition was lawful. The Conseil d'État (France's top administrative court and legal adviser) had warned in 2025 of a possible risk of unconstitutionality tied to property rights.

    Achieved whenthe first object is returned by a decree under the law, published in the Journal officiel, and the decree survives any court challenge.

    leclubdesjuristes.com

    Claimed by

    the French government and Parliament, May 2026, as reported by the Club des juristes

  2. A chatbot for looted art

    Michael Santoro, Haibing Lu and Michele Samorani of Santa Clara University built an AI Provenance Assistant that answers plain-language questions about art that passed through the Jeu de Paume in Paris, where the Nazis gathered works seized in France. "Right now, the tool is only scraping records from one database", and no accuracy figures were given.

    Achieved whenan outside provenance researcher publishes a test of its answers against the original archival records.

    wlrn.org

    Claimed by

    the Santa Clara University team, August 2026, as reported by NPR

Open

  1. 01

    Match collections to looted-art photos

    An image-matching tool that compares every photograph in a museum's online collection with the photographs in the Jeu de Paume database of objects seized by the ERR (the Nazi agency that looted art in occupied France and Belgium), and lists close matches for a provenance researcher to check.

    The brief
    Credit · unclaimed
  2. 02

    Test the red-flag scores

    A published test of tools that score provenance texts for warning signs, such as gaps between 1933 and 1945, words of doubt, and names of dealers linked to looting, run on objects whose history museum researchers have already investigated and judged.

    The brief
    Credit · unclaimed

Open 01 · the brief

Match collections to looted-art photos

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

An image-matching tool that compares every photograph in a museum's online collection with the photographs in the Jeu de Paume database of objects seized by the ERR (the Nazi agency that looted art in occupied France and Belgium), and lists close matches for a provenance researcher to check.

Why it is still open

The database, a joint project of the Claims Conference and the US Holocaust Memorial Museum, covers "more than 20,000 art objects taken from Jews in German-occupied France" and brings together the ERR's surviving registration cards and photographs. The one AI tool reported for it, Santa Clara's assistant, works on the text of its records; Carla Shapreau of UC Berkeley said such tools should link to the original archival records because the database's summaries "are susceptible to errors". No comparison of its photographs with museum collections was found.

archives.govwlrn.org

Built with what exists

An open image-matching model such as DINOv2 or CLIP; open-access images and records from museums such as the Smithsonian, the Met and the Rijksmuseum; a web agent to gather the database's images within its terms of use (see the Note "Deleting 80% to start over"); a review screen for researchers.

  1. First step

    Ask the database's managers about bulk access to its images, and test matching on 20 objects whose ERR history is already known.

  2. Done looks like

    Among 10,000 museum images, a provenance researcher hides 50 objects whose Jeu de Paume history is already published; the tool ranks at least 45 of them in its top ten matches. Then it runs across one museum's full online collection, and every candidate it reports is checked and the results published.

Open 02 · the brief

Test the red-flag scores

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A published test of tools that score provenance texts for warning signs, such as gaps between 1933 and 1945, words of doubt, and names of dealers linked to looting, run on objects whose history museum researchers have already investigated and judged.

Why it is still open

The Looted Art Detector, built at the GLAMhack 2021 hackathon, counts doubtful words such as "probably" and "possibly" and "red flag" names in provenance text to "identify high priority artworks for provenance research"; it says its scores "do not signify that an artwork is looted", and it gives no evaluation of its accuracy.

artdata.pythonanywhere.com

Built with what exists

Provenance texts that museums publish online; the Looted Art Detector's method as a baseline; a language model that answers many fixed questions about each text in one pass (see the Note "A thousand questions, one pass"); an open scoring script.

  1. First step

    Ask one museum's provenance team for 100 researched records with their verdicts, and run the Looted Art Detector's method on them.

  2. Done looks like

    On 500 objects that museum staff have researched and classed as cleared, problematic or unresolved, the best tool ranks a problematic object above a cleared one in at least 90% of pairs, and the test set and results are published so that any tool can be scored on them.