BRENNEN

Industries / Retail, hospitality and personal services / Photography

Stock and licensing

This is selling photographs already taken, through picture libraries and agencies, and the rights and licences that let others use them.

7

Milestones achieved, from 1884 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1884

    Photographs can be copyrighted

    In January 1882 the New York portrait photographer Napoleon Sarony posed Oscar Wilde for a portrait he titled "Oscar Wilde, No. 18", and the Burrow-Giles Lithographic Company printed copies of it. The company argued that a photograph is a mechanical copy of what is in front of the lens, with no originality of thought. On 17 March 1884 the US Supreme Court held that Congress may protect photographs "so far as they are representatives of original intellectual conceptions of the author", pointing to the trial court's finding that Sarony had posed Wilde, chosen the costume and props, and arranged the light and shade, and it affirmed Sarony's $600 judgment.

    LessonThe court found the photographer's authorship in the choices made before the shutter opened, not in the camera.

    law.cornell.edu

    Credit

    Napoleon Sarony; Justice Samuel Miller, who wrote the opinion

  2. 2006

    Pictures for a dollar

    iStockphoto, a community website in Calgary, Canada, where photographers sold pictures to buyers, let buyers license images "for as little as a dollar", a model the trade calls microstock. Getty Images bought it on 9 February 2006 and said its founder, Bruce Livingstone, "pioneered the micropayment stock photography business model".

    LessonGetty answered the cheap market by buying it, to serve, in its own words, "image buyers at all price points".

    sec.gov

    Credit

    Bruce Livingstone, iStockphoto

  3. 2021

    Free photos join Getty

    Unsplash began about 2013 as a Tumblr blog with ten photos. By March 2021 it held more than two million photos that had been downloaded more than three billion times, by its founder Mikael Cho's count, and on 30 March 2021 Getty Images bought it. The UK Competition and Markets Authority (CMA) found in 2026 that about 90% of Unsplash's library is free stock images, and that the main free libraries, Unsplash and Canva's Pexels and Pixabay, are owned by the main paid suppliers.

    LessonGiving photos away became one more line of the paid libraries' business.

    unsplash.comassets.publishing.service.gov.uk

    Credit

    Mikael Cho, Unsplash

  4. 2022

    A small claims board opens

    The CASE Act of 2020 created the Copyright Claims Board, a tribunal inside the US Copyright Office for copyright claims worth up to $30,000, as a cheaper choice than federal court. It opened on 16 June 2022. Many expected photographers to bring most claims; by October 2023 photographs were the largest single type of work, but most infringement claims involved other works, by the Board's account. By March 2026, 1,920 claims had been filed, 36% of them over pictorial, graphic and sculptural works; 787 had been dismissed after the Board's compliance review and 47 had reached a final determination.

    LessonMost claims stop on paperwork before anyone weighs the merits; only 43% of claims filed by the end of 2025 were found compliant, by the Office's count.

    blogs.loc.govccb.govcopyright.gov

    Credit

    US Congress; US Copyright Office

  5. 2023

    Pictures sold as training data

    Shutterstock licenses its images and their captions and other metadata to makers of AI models. In its 2023 annual report it said revenue from content licences fell 7% to $737.3 million and paid downloads fell from 173.3 million to 153.0 million, while growth in its data offering added $84.9 million of revenue. It cited "increased demand for access to our metadata for machine learning and generative artificial intelligence model training".

    LessonIn one year the oldest part of the business, selling downloads, shrank while selling the archive to model makers grew.

    sec.gov

    Credit

    Shutterstock

  6. 2025

    Getty's AI case mostly fails

    Getty Images sued Stability AI in London over Stable Diffusion, an image generator trained on pictures gathered from the internet, which Getty said included millions of its own. Getty dropped its claims about training and about outputs, because the training did not take place in the UK and Stability had blocked the prompts in evidence. On 4 November 2025 Mrs Justice Joanna Smith held that a model "which does not store or reproduce any Copyright Works (and has never done so) is not an 'infringing copy'". She found trade mark infringement only for some Getty and iStock watermarks made by older versions, findings she called "both historic and extremely limited in scope".

    LessonA claim about training could not be won in a country where the training did not happen.

    judiciary.ukjudiciary.uk

    Credit

    Mrs Justice Joanna Smith, High Court of England and Wales

  7. 2026

    Getty drops the Shutterstock deal

    Getty agreed on 6 January 2025 to buy Shutterstock, its closest rival in stock pictures. In May 2026 the CMA found that the deal raised no competition concerns for stock content worldwide, because of competition from Adobe, Canva and AI-generated images, but would reduce choice for UK media buying editorial pictures of news, sport and celebrities. It cleared the deal on condition that Shutterstock's editorial business was sold. On 7 July 2026 Getty ended the merger agreement instead.

    LessonThe regulator counted AI-generated images as a real competitor to stock photographs.

    assets.publishing.service.gov.ukgov.ukgov.uk

    Credit

    UK Competition and Markets Authority; Getty Images; Shutterstock

Watch for

  1. Getty's appeal on AI models

    Getty argues that Stable Diffusion can be an "infringing copy" in UK law even though it holds none of the images it was trained on, because making its model weights involved copying them. The High Court gave permission to appeal on this point in December 2025, and the Court of Appeal heard the appeal on 11 November 2026.

    Achieved whenThe Court of Appeal rules on whether a model trained abroad that stores no copies can be an infringing copy in the UK.

    mlex.comburges-salmon.com

    Claimed by

    Getty Images, November 2026

Open

  1. 01

    Rights data after upload

    A fresh, public test of which services keep or strip the rights information that photographers embed in their image files, run the way IPTC ran its tests and widened to the services that deliver and sell photographers' work. IPTC is the body of news agencies and media companies that sets the standard for that information. Upload IPTC's reference image, which carries a known value in every standard field, including Creator, Copyright Notice, Credit Line, Web Statement of Rights and Data Mining, a field that says whether the owner permits data mining such as AI training. Upload as well a photo signed with Content Credentials (C2PA, a signed record of who made a picture and how it was edited). Then view and download both, at full size and web size, from social and portfolio sites, stock contributor portals and the client galleries photographers deliver through, such as Pixieset, Pic-Time, ShootProof, SmugMug and Zenfolio, and record which fields survive.

    The brief
    Credit · unclaimed
  2. 02

    Claim checks before filing

    A public test, built from past cases at the US Copyright Claims Board, of whether AI models can spot the defects that make the Board send a claim back. For each case, take the claim as first filed and the Board's noncompliance order, which lists what was wrong. Ask each model to list the defects in the original claim, and score its list against the Board's. The test uses past public filings only and gives no advice on live cases.

    The brief
    Credit · unclaimed

Open 01 · the brief

Rights data after upload

Credit · unclaimed

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

What to build

A fresh, public test of which services keep or strip the rights information that photographers embed in their image files, run the way IPTC ran its tests and widened to the services that deliver and sell photographers' work. IPTC is the body of news agencies and media companies that sets the standard for that information. Upload IPTC's reference image, which carries a known value in every standard field, including Creator, Copyright Notice, Credit Line, Web Statement of Rights and Data Mining, a field that says whether the owner permits data mining such as AI training. Upload as well a photo signed with Content Credentials (C2PA, a signed record of who made a picture and how it was edited). Then view and download both, at full size and web size, from social and portfolio sites, stock contributor portals and the client galleries photographers deliver through, such as Pixieset, Pic-Time, ShootProof, SmugMug and Zenfolio, and record which fields survive.

Why it is still open

IPTC tested about a dozen sites in 2013, 2016 and spring 2019; in 2019 it found, for example, that Flickr replaced the embedded Creator with the account owner and that its saved files had no XMP metadata (XMP is a common format for such data). Its lists did not include client-gallery services, and this research found no later round. IPTC's standard now has a Data Mining property, for "data mining prohibition or permission, optionally with constraints", which the past tests did not check.

iptc.orgiptc.orgiptc.org

Built with what exists

IPTC's reference image files and test procedure; ExifTool, a free program that reads and writes image metadata; c2patool, the C2PA project's open-source reader for Content Credentials; free or trial accounts on social, portfolio, stock-contributor, gallery, messaging and cloud-storage services; a results grid like IPTC's.

  1. First step

    Download IPTC's current reference image, upload it to two gallery services and three social sites, download it again, and compare the files with ExifTool.

  2. Done looks like

    At least 20 services, at least 8 of them client galleries, each tested twice a month apart, with the ExifTool and c2patool output for every uploaded and downloaded file published and a grid showing, field by field, whether it was shown, kept or removed.

Open 02 · the brief

Claim checks before filing

Credit · unclaimed

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

What to build

A public test, built from past cases at the US Copyright Claims Board, of whether AI models can spot the defects that make the Board send a claim back. For each case, take the claim as first filed and the Board's noncompliance order, which lists what was wrong. Ask each model to list the defects in the original claim, and score its list against the Board's. The test uses past public filings only and gives no advice on live cases.

Why it is still open

By the Copyright Office's count, only 43% of claims filed by the end of 2025 were found compliant, and the Board spends as much as three-quarters of its time reviewing claims and writing noncompliance orders; the Office says claimants who get help before filing prepare compliant claims more often. Katie Fortney and David Hansen studied the Board's first 880 claims (2024), and K. C. Webb (2026) found from original research that most cases are dismissed, usually for procedural defects, before a decision on the merits. This research found no test of automated checks against the Board's orders.

copyright.govapi.crossref.orgapi.crossref.org

Built with what exists

The Board's public case search at dockets.ccb.gov; the Board's Handbook, for the list of requirements; AI models that read documents, such as Claude Fable 5.2 (see the Note "Three working devices"); a spreadsheet for scoring.

  1. First step

    Pull 20 claims and their noncompliance orders from the docket and write down the defect categories they use.

  2. Done looks like

    At least 200 past claims with their noncompliance orders, a published answer key, and for each model the share of the Board's listed defects it found and the share of its flagged defects that the Board also listed, with a second person checking the key on 50 claims.