HER.01
Collection Cataloguing
Descriptions, transcriptions and Getty vocabulary terms proposed into a review state in TMS, Axiell or CollectiveAccess; a cataloguer accepts each one.
Open pageIndustry
Collections AI,
with a curator on the record.
Curators sign every record. The backlog catalogued, the provenance trail followed source by source, and every visitor guided in their own language. Models propose; curators, registrars and specialists decide what goes on the record.
Last reviewed 30 October 2026 · Engineering guidance, not legal or regulatory advice.
In context
AI in museums means using models to describe, tag and connect collection records, trace provenance through archives and translate interpretation for visitors, while curators and registrars remain responsible for what the record and the gallery say.
Where it breaks today
01
The backlog outlives the staff
Most collections hold objects with a one-line record or none at all. At current cataloguing rates the backlog will not be cleared, and an object nobody can find is an object nobody can study, lend or show.
02
Provenance hides in other archives
The ownership history of a painting is spread across dealer stock books, sale catalogues, letters and restitution databases, in several languages, and a gap in the 1933–1945 period is a legal and ethical question, not a footnote.
03
Interpretation in one language
Labels and audio guides exist in two or three languages at best, and updating them for a new display means a new translation round and a new recording.
OUR APPROACH
AI for museums that respects the record
A collection record is a scholarly and legal document. So the systems we build propose and cite; they never write to the record on their own. A suggested description is marked as machine-generated until a cataloguer accepts it. A provenance link carries the archive reference it came from. A visitor guide speaks from curatorial text, not from what a general model happens to believe about an artist.
We work inside the collection management system you already run, such as TMS, Axiell Collections or CollectiveAccess, and with the standards it already follows: Spectrum procedures, Getty vocabularies (AAT, ULAN, TGN), LIDO and CIDOC CRM for exchange, and IIIF for images.
Provenance and the art market carry real legal weight. The Washington Conference Principles on Nazi-confiscated art, the 1970 UNESCO Convention, the EU import rules for cultural goods (Regulation 2019/880) and anti-money-laundering duties on art market participants under the EU AML directives and the UK Money Laundering Regulations all shape what due diligence means. We build research tools; conclusions and compliance decisions stay with your specialists and legal advisers.
HOW IT DEPLOYS
PHASE 01
Sample the collection
A few hundred records chosen by the registrar, including the hard ones, to see what the models get right and wrong on your material.
PHASE 02
Connect the CMS
Suggestions written to a review state in your collection system, mapped to your fields and vocabularies, never directly to the live record.
PHASE 03
Curators accept, at scale
Review queues sorted by confidence, so specialist time goes on the uncertain records and every acceptance is logged by name.
What we build
01
Collection Cataloguing
Descriptions, keywords and controlled-vocabulary terms proposed from images, old cards and ledgers, for cataloguers to accept or correct.
SEE COLLECTION CATALOGUING →02
Provenance Research
Search across catalogues, archives and loss databases for each owner and sale in an object’s history, with every lead cited for a researcher to verify.
SEE PROVENANCE RESEARCH →03
Multilingual Guides
An AI tour guide that answers visitors in their own language from curatorial text, by phone, QR code or audio, with accessibility built in.
SEE MULTILINGUAL GUIDES →04
Search Across the Collection
Retrieval over records, archives and research files, so staff and the public can search by meaning across everything the institution holds.
SEE RAG SYSTEMS →Controls we implement
MACHINE TEXT MARKED UNTIL ACCEPTED
SOURCE CITED ON EVERY LEAD
NO DIRECT WRITES TO THE RECORD
ACCEPTANCE LOGGED BY NAME
LENDER AND DONOR DATA RESTRICTED
NO TRAINING ON YOUR COLLECTION
These are engineering controls we build into the architecture, not certifications. QQuantum.ai holds no security certifications, and your due-diligence and legal obligations remain yours — see our trust centre.
Museums, heritage & auction use cases
-
-
HER.02
Provenance Research
Searching and reading across sale catalogues, archives and loss databases, every lead cited and every gap flagged; researchers confirm each step and draw conclusions.
Open page -
HER.03
Multilingual Guides
A browser-based AI tour guide answering visitors in their language from curatorial text only, by text or voice, with access modes built to WCAG 2.2 AA.
Open page
Questions
Questions about Museums, Heritage & Auction.
How are museums using AI?
In three main ways. Behind the scenes, computer vision and language models describe and tag objects, transcribe old catalogue cards and connect records, which makes collections searchable. In research, they help trace provenance and read archives. In the galleries, they power multilingual guides and conversational interpretation. The careful versions keep curators responsible for what the record and the gallery say.
Can AI catalogue a museum collection?
It can draft much of the work: a physical description from photographs, keywords mapped to the Getty Art and Architecture Thesaurus, transcriptions of handwritten cards and ledgers. It cannot attribute, date or assess significance with authority. We write suggestions to a review state in your collection management system, marked as machine-generated, for a cataloguer to accept or correct.
Can AI help with provenance research?
Yes, as a research assistant. It can search digitised sale catalogues, dealer records and loss databases for names and objects, read documents in several languages, and assemble a timeline with a citation for each step. Gaps, especially in the 1933–1945 period, are flagged rather than filled. Conclusions about ownership and any restitution question remain with provenance specialists and the institution.
How do auction houses use AI?
Mostly for cataloguing lots at speed, searching past sale records, checking objects against stolen and looted art databases, and supporting due diligence on buyers and sellers, including the anti-money-laundering checks art market participants must carry out. Attribution, valuation and the decision to accept a consignment stay with specialists and compliance staff.
Is AI-generated interpretation accurate?
Only if it is grounded. A general model will confidently invent facts about lesser-known objects. Our guides answer from the institution’s own curatorial text and records, say when they do not know, and are reviewed by curators before launch and sampled afterwards. Visitors are told they are talking to an AI.
