Industries / Retail, hospitality and personal services / Retail and e-commerce
Auction houses
Auction houses sell other people's art, collectibles, property and equipment to the highest bidder in an open, timed contest, and are paid by fees taken from the buyer, the seller or both.
1674
Baron Claes Rålamb, governor-general of Stockholm, issued the ordinance that founded the city's auction chamber, today Stockholms Auktionsverk, which describes itself as the world's oldest auction house.
35
Milestones achieved across 6 domains.
11
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Auctions with printed catalogues and numbered lots began with book sales in the Dutch Republic, and the earliest catalogue that survives is for a library sold in Leiden in 1599. Stockholm's state auction chamber opened in 1674, Samuel Baker held the first sale of what became Sotheby's in London in 1744, and James Christie held his first in 1766. Sotheby's black-tie evening sale of 1958 turned the top auctions into public events, and from 1975 the London houses charged buyers a fee as well as sellers.
Where it stands
Auction houses sold an estimated $24.8 billion of art in 2025, counting public auctions and private sales together, by Arts Economics' count for Art Basel and UBS; public auctions rose 9% to $20.7 billion, while online-only sales across the whole art market fell to 15% of the total, from 25% in 2020. The same contest sells cars and machines: Copart moved all its salvage vehicle auctions to the internet in 2003, and Ritchie Bros. moved its equipment auctions online in 96 hours in 2020. At the top of the art market many lots are backed by guarantees before the sale, and in the UK and the EU houses must check who is buying and selling under anti-money laundering law.
What is open
Much of what sets a price still happens out of view: who financed a lot, how many people bid, and what the house charged the seller. Several open jobs below turn public catalogues, results and sale videos into records anyone can check, such as a ledger of guaranteed lots and a count of bidders taken from livestreams. Others are independent tests of tools now sold to auction houses: AI that writes catalogue entries, judges whether a painting is genuine, screens ownership histories for looted art or reports a car's condition, none of which has accuracy figures checked by an outsider. AI models that read photos, handwriting and speech, and agents that collect pages from public websites, put this work within reach of a student or a small team.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Taking in goods
Taking in goods is winning consignments, the objects owners hand to a house to sell, and agreeing the terms: the estimate, the reserve (the secret lowest price the seller will accept), the seller's commission and, at the top of the market, a guarantee of a minimum price.
6 · 1 · 2
Last achievedA fixed seller fee reversed2025
Watch forIntake estimates from photos
2
openValuation and cataloguing
Valuation and cataloguing is working out what a lot is and what it may fetch, and writing it up in the catalogue: the lot number, title, maker, date, material, size, condition, ownership history and a pre-sale estimate, given as a low-to-high price range.
4 · 2 · 2
Last achievedAn auction house buys AI2018
Watch forAI-written lot descriptions
2
openAuthenticity and ownership history
This is checking that a lot is what the catalogue says it is and that the seller has the right to sell it: that it is not a fake, not stolen, not looted from a person or a country, and not exported against the law.
6 · 3 · 2
Last achievedLab tests unmask a "Hals"2016
Watch forAI attributions of paintings
2
openThe sale
The sale is the auction itself: the auctioneer takes bids from the room, the telephones and the internet, in steps set by the house, until one bidder is left and the hammer falls.
6 · 1 · 2
Last achievedBids from screens worldwide2020
Watch forBids read from sale videos
2
openOnline bidding
Online bidding is taking bids over the internet, either during a live sale or in online-only sales that run for days and close at set times, now the usual route for cars, machines and lower-priced art.
7 · 2 · 2
Last achievedA data theft before sales2024
Watch forLivestream auctions at scale
2
openPayment and shipping
After the hammer, the buyer pays the hammer price plus the buyer's premium and taxes, the house checks who is paying and where the money comes from, pays the seller after taking its commission, and the lot is packed, insured, licensed for export where needed and shipped.
6 · 2 · 2
Last achievedImport papers for old goods2025
Watch forOne EU money-laundering rulebook
2
open
Advances to build with
2 research notes

Claude Fable 5.2
Three working devices
Anthropic says its AI model Claude Fable 5.2 designed three prototype devices that worked without electrical modifications, with no outside test reported.

Microsoft ThinkingBox
All twenty times
By Microsoft's own count, 50 steps of training took an AI model from passing 38 of 507 test tasks on all 20 tries to passing 107.
Open in auction houses
3 of 12 open deliverables
Open · Taking in goods
Public table of seller terms
A free table of what 100 auction houses in at least 10 countries charge sellers: commission rates by price band, fees for photography, insurance and unsold lots, any success fee, whether they offer guarantees, and how soon they pay, each with a link to the house's published terms and the date checked.
Built with what exists
Houses' public terms-of-business pages, collected with Browser Use 1.0 where each site's terms allow (see the Note "Deleting 80% to start over"); Claude Fable 5.2 to read each page into fixed fields (see the Note "Three working devices"), with every row checked by a person; a spreadsheet or small database with a public log of changes.
Open · Valuation and cataloguing
Audit AI catalogue entries
A blind test of AI cataloguing tools on 200 objects whose facts are already settled: each tool writes an entry from the photos, and two checkers count factual errors in maker, date, material, size and stated condition.
Built with what exists
Objects with photos and checked records from museum open-access collections; the commercial tools themselves (Gavelist charges $0.15 a lot); Claude Fable 5.2 as a baseline (see the Note "Three working devices"), with each tool run several times on the same lot to show how much its answers vary (see the Note "All twenty times"); a scoring sheet with two checkers per entry.
Open · Authenticity and ownership history
Blind test of AI authentication
A sealed set of images of 100 paintings and drawings whose status is settled, mixing works with secure attributions, documented forgeries and old copies, sent to every AI authentication service that will take them, with the answer key kept from the services until all results are in.
Built with what exists
Museum open-access images for secure works; published images of exposed forgeries, such as the Beltracchi fakes and the "Hals" portrait; the services' own submission routes; the answer key stored with a timestamp in a public repository before any result comes back.
Where it sits
Broader