AI Licensing Moved to Marketplaces. Most Publishers Stayed in Court.

In the second week of March 2026, two things happened in the AI copyright space. 󠇟󠇠󠇡󠇢󠇯󠆚󠇂󠅀󠇁󠄦󠄬󠄬󠅮󠆹︄󠇗󠅬󠇊󠆉︍󠄗󠅅󠆒󠄥󠆔󠇐󠆃󠇬󠆯󠄀󠄵󠅮󠅕󠆈︎󠆃󠄯󠅳󠄗󠄑󠄞󠄧󠄟󠇐Britannica and Merriam-Webster 󠇟󠇠󠇡󠇢󠆝󠇜󠄄󠆨󠇗󠇄󠅱󠅮󠅉󠆣󠅿󠆤󠆼󠇠󠄇󠇮󠆆󠄼󠇄󠇇󠇙󠆐󠄡󠇌󠄰󠆖󠅞󠅻󠅌󠅄󠄨󠅒󠇨︉󠄅󠇞󠆽󠅦󠅡󠆹filed suit against OpenAI in the Southern District of New York, alleging the company used nearly 100,000 copyrighted articles to train ChatGPT without permission. 󠇟󠇠󠇡󠇢󠇫󠇎󠆐󠆊󠇊󠅤󠄞󠇡󠅫󠆁󠄶󠄴󠇑󠄗󠆓󠇈󠆶󠄽󠆂󠄀︃󠆞󠄝󠄋󠄏󠅳󠅹︄󠄁󠄚󠅯󠄸󠄒󠇎󠆠󠄡󠅋󠄱󠄾󠆍The same month, the News/Media Alliance signed a licensing deal with AI startup Bria that gives 2,200 publisher members a path to recurring revenue from AI retrieval. 󠇟󠇠󠇡󠇢󠄖󠅍󠆂󠄇󠇒󠆮󠇖︀󠅨󠅘󠇏󠇊󠅸󠄲󠄏󠇥󠄕󠇬󠄨︊󠄾󠆁󠄫󠆤󠄂󠇦󠆖󠅃󠄝󠇔󠆍󠅬󠆚󠅜󠄪󠇓󠆤󠇭󠆑︄One event generated headlines. 󠇟󠇠󠇡󠇢󠆄󠄴󠆲󠆕󠆄󠅂󠄻︊󠅺󠇜󠄽󠅟󠄎󠅹󠆶󠆇󠄆󠅀󠆃󠅗󠇭󠄭󠇈󠆳󠅢󠆸󠇁󠆋󠄕󠇪︇󠅘󠄞󠇐󠇢󠇍󠆗󠅣󠆳󠅝The other generated revenue. 󠇟󠇠󠇡󠇢󠇪󠄻󠆱󠆿󠅃︍︂󠅷󠆽󠇃󠅔󠆬󠄡󠄥󠅫󠅤󠆧󠆞󠆴󠇩󠆫󠄇󠄁󠄜󠆒󠄓󠆵󠅡󠅤󠅊󠇀󠆫󠇫󠅉󠄯󠇎󠅴󠅔󠄦󠇩The licensing war for AI-used content has moved from courtrooms to marketplace infrastructure, and publishers without machine-readable provenance signals are invisible to the systems that would pay them.

󠇟󠇠󠇡󠇢󠅚󠆠󠆭󠇄󠆡󠄋󠅔󠅅󠄈󠄓︋󠄻󠆠󠇫󠆝󠆢󠇚󠆪󠄫󠆀󠅖︍󠄁󠆑󠇆󠅪󠅐󠇭󠅾󠅎󠅣󠆠︋󠇍󠅞󠅶󠅅󠆹󠄶󠇑This post discusses legal developments for informational purposes only and does not constitute legal advice. 󠇟󠇠󠇡󠇢󠆆󠄉󠄌︁󠆯󠅋󠅲󠄸󠅍󠄞󠇚󠅕󠅌󠅖󠅖󠆧󠇟󠅫󠇭󠆫󠄶󠄀󠆊󠆉󠅖󠆌󠆏󠅕󠅴󠇇󠆸󠇇󠇆󠇜󠅮󠇫󠄚󠄓󠇇︁Encypher is a technology company, not a law firm. 󠇟󠇠󠇡󠇢󠇉︊󠆾󠇥󠇆󠄕󠄼󠆐󠆁󠅔︃󠆀󠅕󠇪󠄦󠅳︂󠆙󠆍︃󠇘󠅔󠆅󠄎󠇔󠄙󠆳󠄎󠅟︃︉󠆽󠄺󠅟󠅲󠅷󠆫︎󠆻󠅟Consult qualified legal counsel for advice specific to your situation.󠇟󠇠󠇡󠇢󠅭󠄤󠅮󠆌󠇎󠄯︈󠆉󠆥󠆻󠄿󠇚󠅆󠆱󠆟󠅃󠆕󠆭󠆆󠄲󠅉󠅞󠇧󠆹󠇩󠆸󠅝󠆵󠄊󠅤󠆣󠆘󠅤󠄖︅󠄃󠆺󠇐󠅈󠅤

The Marketplace Moment

AI companies stopped just scraping content and started shopping for it.

󠇟󠇠󠇡󠇢󠅈󠄰󠄃󠄑󠅖󠄅󠆮󠇜󠅰󠅝️󠄮󠅥󠅚󠆃︀󠄬󠅗󠆲󠇨︋󠆧󠅪󠆠󠇪󠇭󠄮󠄴︅󠇎󠆐󠇍󠅏󠆆󠄗󠆷󠆂󠆎󠄌󠆞On February 3, Microsoft launched the Publisher Content Marketplace - a licensing hub where publishers set usage terms and AI systems discover and license content for grounding scenarios. 󠇟󠇠󠇡󠇢󠄈󠅣󠅪󠄫󠅊󠅶󠅟󠆥󠆽󠆋󠇉󠄋󠅾󠆀󠇐󠇊󠇂󠅚󠇘󠅹︌󠇋󠅥󠄥󠄽󠇕󠄩󠆄󠅈󠄙󠆽󠇠󠄉󠆋󠇎󠆒󠇫󠄘󠄮󠅸The launch partners include the Associated Press, Conde Nast, Hearst Magazines, and Vox Media. 󠇟󠇠󠇡󠇢󠆌󠅙󠅜󠅮󠄑󠄛󠅦󠆥︊󠅟󠆀󠄋󠆹󠆛󠇄󠄝󠆈󠄃󠆁󠆻󠅛󠅜󠆤󠅸󠄃󠅹󠆋󠇘︋︄󠅻󠄣󠇇󠄏󠇋󠅬󠅦󠇤󠆼󠆾The platform provides usage-based reporting that gives publishers visibility into how their content creates value in AI pipelines. 󠇟󠇠󠇡󠇢󠇈︉󠄈󠆺︆󠅬󠅘󠆀󠇝󠄼󠄸󠇅󠆓󠄄󠇧󠇇󠄸󠇣󠄺󠅮󠄗󠅎󠄧︈󠇢︃󠄦󠄖󠆥󠆊󠄭󠇏󠄐󠆫󠅠󠅘󠇒󠆽󠄭󠅎Microsoft describes it as building toward a "sustainable content economy for the agentic web."

󠇟󠇠󠇡󠇢󠆒󠇕󠄊︋󠆐󠇒󠄆󠅡󠅽󠆅󠆟󠆃󠅴︆󠆭󠄠︁󠅅󠄕󠄌󠄻󠄙󠇑󠆼󠅘󠄍󠅞󠅞󠄥󠅌󠄫󠅎󠄟󠄄󠅶󠄜󠇑︎󠆷󠆪The NMA/Bria deal operates on a different model but with the same prerequisite. 󠇟󠇠󠇡󠇢󠇆󠇥︎︁󠄳󠇉󠇦󠅯󠆥󠄉󠅎󠄘󠇖󠄢󠆅󠇖󠇍󠇭󠇪︌󠄾󠇏︎󠆩󠆍󠇮󠄅󠄜󠅡󠅾󠆸︋󠇛󠆫󠄩󠇃󠄁󠄴󠄄󠄟Revenue splits 50-50 between Bria and the publisher based on an attribution model. 󠇟󠇠󠇡󠇢󠆧󠅤󠅫󠅠󠇞󠅸󠆸︌󠇃󠅗󠄠󠄷󠄛󠅍󠅄󠄈󠇩󠄵󠅐󠇁󠄍󠅫󠆒󠆏󠇙󠆰󠇚󠇠️󠄬󠄼󠄩󠅞󠇅󠆶️️󠄸󠄘󠆝For the 2,200 member publishers - many of them local and regional outlets without the resources to negotiate directly with OpenAI or Google - this is the first viable channel for monetizing RAG-driven enterprise demand.

󠇟󠇠󠇡󠇢󠇋󠆝︍󠅉󠄜󠆬󠅮󠅮󠄠︆󠄷󠅧︃󠅋󠇙󠅹󠄎󠆙󠇧󠆺󠆥󠄿󠆑󠄪󠆌󠇘󠅢󠅝󠄎󠆅󠄜󠅩󠄃󠆿︇󠄡󠄲󠆳︂󠆫These are not isolated experiments. 󠇟󠇠󠇡󠇢󠆰󠄹󠆃󠅊󠅦󠆕󠅬󠄨󠅚󠆁󠅊󠅴󠄟󠅑󠇂󠅬󠇒󠅶󠄀󠄞󠇍󠅰︉󠅔󠄯󠇀󠇋󠆴󠇎󠄭󠄢󠆃󠄥󠇕󠆒󠅻󠆾󠆙󠇡󠆕They represent a structural change in how AI companies acquire content rights. 󠇟󠇠󠇡󠇢󠇆󠇁︄󠆍󠅥󠄏󠆐󠄤󠄙󠆮󠄳󠆱󠅽󠅙󠆲󠄑󠅍󠆂󠅮󠅒󠇙󠇧󠄖󠇢󠅾󠇀󠅏󠆃󠆂󠇑󠆿󠅳󠆱󠄹󠅁󠆄︈󠅔󠄂󠇥The economics are straightforward: as retrieval-augmented generation becomes the dominant architecture for enterprise AI, the companies building these systems need licensed content they can point to when a customer's legal team asks where the grounding data came from. 󠇟󠇠󠇡󠇢󠇁󠅉󠆬󠆓󠇦󠄹󠅫󠅺󠄂󠄍󠇛󠄟󠆽󠄤󠇧󠄻󠆕󠆛󠅽󠆸󠆘󠄂󠅳󠇬󠄠︊󠅁󠅺󠇟󠆡󠄀󠆩󠆻󠆖󠆿󠅧󠅇󠄌󠅈︉The demand side of the licensing market now exists. 󠇟󠇠󠇡󠇢󠅱󠅒︍󠇐󠅙󠄨󠇔󠆬󠅫󠆾󠄲︎󠇑󠄤󠇍󠆙󠅝󠇔󠇗󠄀󠆤󠆨󠆀︀󠅘󠄺󠅱󠇘󠅒󠅩󠄮󠆩󠅁󠄁󠆭󠆼󠆯󠇤󠄣󠄓The constraint is the supply side.

󠇟󠇠󠇡󠇢󠅜󠆿󠅗󠄐󠄯󠅽󠆣󠆈󠅲󠆰󠆶󠇬󠅙󠄟󠄋󠅒󠄳󠄇󠆖󠇌󠄘󠆺󠆱󠆅󠆗󠇋󠇬󠆉󠇋󠇌󠆓󠅺󠆜󠇦󠆇󠅪󠄥󠇉󠆽︃Every one of these marketplaces requires the same thing from publishers: content that can be programmatically identified, attributed, and tracked. 󠇟󠇠󠇡󠇢󠇘󠇋󠄈󠄖󠇞󠅥󠇓󠄻󠄁︊󠅧󠆼󠅊󠆡︋󠅋󠆻󠅞󠅠󠅓󠅃󠄙󠄣󠄥󠄯󠄠󠆏󠄁󠄙󠅅󠄖󠄎󠅯󠆣󠅃󠆠󠄮󠆬󠄋󠆌Microsoft's PCM requires publishers to structure metadata and archives so content is discoverable. 󠇟󠇠󠇡󠇢󠅓󠆃󠆥󠆄󠆘󠅳󠆶󠇋󠆽󠇯󠅜󠇤󠆾󠄗󠄥󠆼︉󠅪󠄋󠄑󠆅󠆿󠆈󠄔︋󠄲󠅘󠅔󠆌󠅾󠇭󠄪󠄇󠅮󠆑󠄚󠅾󠆴󠄭󠇧Bria's attribution model requires content to be identifiable for usage tracking. 󠇟󠇠󠇡󠇢󠆴󠆼󠆌︂󠆫󠅻󠅜󠅀󠄰󠅨󠄎󠄷󠄖󠅣󠇘󠇩󠆏󠇢󠇄󠅙󠄛󠆨󠇜󠄑󠄨󠄏󠄚󠆢󠆹󠄯󠆾󠆒󠄅󠄰󠅄󠄅󠄾︉󠄉󠅎The common infrastructure requirement is machine-readable provenance - signals embedded in or attached to content that allow automated systems to determine who created it, under what terms it can be used, and how to route payment.

󠇟󠇠󠇡󠇢󠆢󠅋󠅰󠅇󠅵󠆂󠇨󠄏︊󠇌󠄩󠅊󠇬󠅥󠆆󠇯󠅒󠅢󠅭︋󠆶󠄰󠄳󠇧󠆦󠆃󠆄󠅃️︉󠅕󠄑󠇜󠇊󠅞󠆽󠅤󠆕󠆥󠆎The common requirement across every platform is the same: machine-readable provenance signals that make content visible to the systems allocating revenue.

󠇟󠇠󠇡󠇢󠅬󠆵󠆨󠄝︋󠆭󠆊󠅏󠆯󠆡󠄶󠄺󠆮󠄯󠅦󠅿󠇃󠄸󠆄󠄭󠅦󠄇︁󠆅󠅜󠆚󠆶󠆉󠆳󠅲󠄿󠆡󠇫󠄦󠇜󠆤󠇍󠅚󠆤󠅂󠇟󠇠󠇡󠇢󠆆󠇥󠇛󠄆󠅍󠅷󠅝󠄢󠇏󠅲󠇊󠇑󠅶󠆐󠇎󠅚󠆵󠅁󠅋󠄪󠆷󠄣󠅛󠅜󠇘󠄊︄󠇮󠅒󠇒󠅦󠄗󠇣󠅘󠄡󠄠󠅣󠇝󠄃󠄗The Infrastructure Gap

The parallel to early web monetization is instructive. 󠇟󠇠󠇡󠇢󠄽󠇄󠆥︌󠄽󠅬󠅢󠇭󠄣︋󠄿󠄑󠇦󠆣︄󠄮󠆖󠇢󠅱󠆾󠅆󠇃󠆖︆󠄾󠄟󠅆󠇌󠄿󠅆󠅴󠆪󠆑󠅠️󠅄󠅌󠆀󠄅󠄆In the mid-2000s, publishers who adopted analytics tags and ad network integrations captured digital advertising revenue. 󠇟󠇠󠇡󠇢󠄭︅󠆴󠆢󠇘󠅅󠅔󠆑󠅺󠆪󠅳󠅻󠇣︎󠄆󠇁󠄫󠇛󠄿󠆟󠄆󠅡󠄞󠄣󠇃󠄠󠆤󠅇󠇀󠅽󠇬󠅡󠇡󠄚󠅡󠇀󠆌︍󠆣󠇦Those who published content without measurement infrastructure were functionally invisible to the systems allocating ad spend. 󠇟󠇠󠇡󠇢󠄞󠆄︀󠄡󠅕󠅥󠅓󠄝󠅏︅󠅫󠆤󠇁󠅦󠆴󠅷󠄍󠄭󠆳󠄡󠇮󠅗󠇜󠆆󠅊󠆅󠇋󠇤󠄠︂󠇦󠅽󠆕󠇚󠄣󠅓󠆏󠄋󠅠󠇇The content existed. 󠇟󠇠󠇡󠇢󠄢󠆦󠄘󠅃󠇅󠇮󠆨󠅑󠄍󠅉󠆿󠆚󠅾󠇫󠅏󠄕󠅉󠇕︀󠅏󠅛󠄈󠇥︈󠄜󠇬󠄽󠆼󠄼󠄉󠇁󠇫󠇟󠄨󠄥󠆑󠆸󠆨󠆔󠇧The audience existed. 󠇟󠇠󠇡󠇢󠅹󠆻󠅳󠆣︋󠅴󠆳󠆞󠄷󠇪󠅱󠄉󠆜󠆷󠇉󠄱󠄀󠅪︉󠄡󠄁󠆀󠅒󠇤󠄏󠆡󠅽︀󠅼󠄻󠆩󠅚󠅒󠇫󠆾󠇢󠄣󠆠󠅗󠅥The revenue went elsewhere because the plumbing was missing.

󠇟󠇠󠇡󠇢󠅪󠄬󠇜󠄺󠄟󠆆󠅳󠄢󠅛󠅴󠇂󠇙󠆳󠅬󠄣󠄣󠆼󠇢︈󠄊󠇀󠄏︄󠆣󠅫󠆣󠄅󠅕󠅉︍󠅹󠆝󠄹󠇜󠅠󠆭󠇒󠄈󠆋󠄆The same dynamic is forming in AI licensing. 󠇟󠇠󠇡󠇢󠄗󠅄󠄦󠄗󠆵󠆫󠄔󠄞󠅙󠆜󠆙󠅲󠅠󠄛󠆙󠄼󠄟󠅵󠄷󠅝󠇑󠄘󠆽󠅯󠄟󠇋󠆧󠅤󠇧󠅏󠅴󠅱︉󠆟󠇤󠆠󠄍󠄔󠇂󠄞The marketplace platforms are the new ad networks. 󠇟󠇠󠇡󠇢󠇐󠄝︃󠆹󠄲󠅄󠄝󠄚󠇟︅󠅅󠄖󠇇󠅬󠅛󠅬󠄺󠄖󠄮󠄀󠅤󠅀󠇧󠄌󠅶󠄣󠅁󠅸󠄚󠅝󠄆󠅨󠆠󠅊󠄉󠆲󠄘󠄆󠇍󠄙Machine-readable provenance is the new analytics tag. 󠇟󠇠󠇡󠇢󠅞󠄭󠅟󠄰󠅂󠇑󠄠󠆕󠅷󠄉󠆰󠅫󠇇󠄡󠅙󠄹︍󠅩󠇕󠆤󠇖󠅁󠅄︋󠄾󠇯󠄐󠄋󠆙󠅏󠆦󠆧󠇧󠅫󠄘󠆕󠄝󠆔󠄢󠆣And the publishers building this infrastructure now are positioning themselves to capture revenue that will flow whether or not the litigation resolves in their favor.

󠇟󠇠󠇡󠇢󠄫󠅛󠆌󠇬󠅵󠄹󠄼󠆢󠅓󠇈󠄦󠇐󠄗󠆠󠆑󠇣󠄽󠇁󠅚󠆆󠇗󠄛󠇚󠄩󠇝︊󠄷󠅴󠆿󠄊󠅙󠆂󠆑󠆣󠆿󠄼󠇖󠄅󠆢️C2PA - the Coalition for Content Provenance and Authenticity - provides the standard for embedding authentication manifests into content. 󠇟󠇠󠇡󠇢󠄶󠇑󠄸󠇎󠆛󠆿󠇒󠄧󠇍󠇒󠅾󠇩󠆽󠇭󠆾󠄭󠇧󠆼󠄰󠅵󠆑󠆳󠇥︀󠇯︎󠅝󠅝󠅊󠅛󠄻󠄐󠇃󠆆󠄹󠇊󠄃󠅄︍󠅅The specification, developed by Adobe, Microsoft, Google, OpenAI, the BBC, the New York Times, and others through c2pa.org, enables a document-level manifest that binds authorship, creation metadata, and usage rights to a piece of content in a cryptographically verifiable format. 󠇟󠇠󠇡󠇢󠆋︌󠆽󠆦󠅤󠅳󠇏󠅺󠇈󠄞󠇌󠇔󠆬󠄻󠅟󠅇󠄖󠄴󠆎󠆥󠄀󠆪󠅘󠅡󠅍󠄸󠅭󠆒󠇛󠆓️︀󠅌󠄽󠇭󠆲︃󠅌󠄐󠅙Section A.7 of the C2PA 2.3 specification, published January 8, 2026, extends this capability to text content for the first time. 󠇟󠇠󠇡󠇢󠅜󠄇󠅌󠅯󠄹󠅵󠆶󠅨󠄴󠄷󠆖󠆲󠅒󠆻󠆿󠄭󠆙󠆶󠇁󠆫󠅼󠆴󠇁󠄓󠅙󠆀󠇛󠄋󠇤󠇧󠄳󠄨󠄥󠅸󠇦󠅬󠆧󠄗󠄌󠅓We co-authored that section, and it represents the first interoperable standard for text provenance.

󠇟󠇠󠇡󠇢󠅣󠅸󠄿󠆘󠆱󠅙󠅫󠄢󠆱󠆂󠅭󠆩󠄢󠄚󠆺󠆣󠄮󠄂󠇨󠇏󠆻󠇋󠅙󠅩󠄑󠆭󠄸󠄖󠆶󠄏󠄘󠄡󠄂󠅂󠄃󠆵󠅮󠆽󠅴󠆃But document-level provenance is only the foundation. 󠇟󠇠󠇡󠇢󠆗󠇚󠇒󠆫󠆝󠄁󠆔︈︋󠄈󠅍󠅒󠆺󠆲󠇤󠅾󠅰󠅠󠇄󠇇󠅅󠇣󠇉󠄞󠆤󠄭󠇫󠄰󠆐󠆜󠅒󠆿󠅭󠇋󠇃󠆐󠅥󠇀󠆮󠆕When a RAG system retrieves three paragraphs from a 5,000-word article and uses them to ground a response, the licensing question is not about the article as a whole. 󠇟󠇠󠇡󠇢󠄺󠆅󠄼󠅭󠅂󠅆󠄰󠆽󠅐󠄬󠅮󠅏󠅌󠆘󠇪󠆷󠅨󠆦󠅌󠅂󠆜󠆼󠆝󠅠󠄫󠄐󠅛󠇥󠄚󠅬󠅪󠅦󠅾󠅃󠄒󠇂󠆑󠇛󠅎󠄸It is about which specific segments were used and under what terms. 󠇟󠇠󠇡󠇢󠄘︋󠇥󠄺󠄥󠇚󠆇󠆍󠆉󠅏󠆋󠇙󠇍󠄵󠇁󠄬󠄻󠆷󠅚󠄥󠆉󠆒󠆌󠆀󠇌︁󠅦󠄻󠅳󠆫󠆤󠅫󠇂󠅷󠄊󠇖󠅯󠄔󠇄󠅍Encypher's proprietary technology enables sentence-level attribution - binding authorship and licensing terms to individual text segments so that provenance survives extraction, redistribution, and transformation. 󠇟󠇠󠇡󠇢󠄃󠆏󠇛󠄜󠄚󠆌󠇖󠄤󠄁󠆜󠄇󠄯󠅴󠅙󠇭󠆴󠆫󠆲󠄡󠄣󠅺󠄏󠄍󠅕󠄼󠆵󠇪󠄾󠆿󠆏󠄬󠇬󠆣󠆂󠇉󠇑󠇗󠅭󠆌󠄺This granularity is what turns a provenance standard into a licensing mechanism. 󠇟󠇠󠇡󠇢󠄦︍󠇕󠅴󠆳󠅃󠄑󠇮󠄺󠆍󠅧󠅝󠇢󠅬︋󠄡󠅆︈󠇜󠇅󠇐󠄥󠆱󠇪︂󠅥󠆯󠄧󠅙󠅦󠅀󠆭󠄃󠄳󠄴󠅹󠅹󠆷󠅽󠆦A marketplace can identify not just that content came from a particular publisher, but which sentences were retrieved, how they were used, and what the publisher's terms require for that specific use.

󠇟󠇠󠇡󠇢󠆺󠅄󠇯󠆜󠆎︇󠆲󠆞󠅴︅󠆺️󠅂󠄝󠆺󠆜󠄃󠆫󠄍󠅎󠆬󠆿󠇐󠆃󠅿󠆺󠄞󠄑󠄣󠄥︆󠅜󠇢󠅘󠄎󠆾󠄹󠄩󠅧󠇕󠇟󠇠󠇡󠇢󠆥󠅈󠄰󠆲󠄉󠆲󠅬󠄓󠆲󠆐󠇗󠅺󠇩󠅑󠇔󠄰󠅰󠄷󠇩󠅘󠅙󠆚󠄊󠄈󠇮󠅁󠄡󠄛󠆘󠄖󠄦︀󠅘󠄰󠇏󠅩󠄔󠇕󠄎󠄃The Litigation Counterargument

The case for continued litigation is not frivolous. 󠇟󠇠󠇡󠇢󠆗󠄠︀󠇔󠅅󠅘󠅸󠄘󠄑󠄪󠄶󠄼󠅔󠅥󠆽󠄵󠆥󠇐󠄅︀󠆇󠆓󠄲󠆍󠄲󠅵󠅳󠄭󠇖󠄏󠆟︀󠄔󠅒󠄥󠅟󠄀󠅬󠆦󠄔It has produced real results.

󠇟󠇠󠇡󠇢󠄤︌󠄗󠆁󠆘󠇃󠅼󠅙󠅘︉󠄀︍󠆭󠆡󠅩󠆆󠆹󠆺󠇓󠆛󠆥󠅂󠆉󠇞󠅿󠆏󠆌󠅆󠅋󠇗󠅍󠇟󠆥󠅑󠄩󠄬󠅵󠇬󠆑󠅹Britannica's suit against OpenAI targets both training-data use and real-time RAG retrieval that reproduces content verbatim. 󠇟󠇠󠇡󠇢󠄻󠆍󠄘󠅢󠇕󠆼󠄃󠄲󠆋󠄷󠆠󠆄󠇅󠄽󠄿󠇣󠅪󠅟󠅟󠆛󠅾󠄳󠆁󠇌󠆔󠅏󠄀󠆟󠆇󠄱󠆼︊󠅌󠄋󠇇󠇠󠆘󠆕󠅬󠄬It includes Lanham Act trademark claims for AI hallucinations falsely attributed to the plaintiffs - a novel theory that expands the legal surface area beyond copyright alone. 󠇟󠇠󠇡󠇢󠆯󠅮󠅷󠅎󠄠󠆪󠅝󠄝󠆥︌󠆎󠅬󠆖󠅦󠅑󠅄󠆭󠆄󠆸󠄢󠄐󠅦󠄩󠆪︂󠇄󠆎󠇋󠄃󠆄󠇋󠇘󠄒󠇤󠄅󠅨󠅕󠆌󠇟󠇐OpenAI responded that its models are "trained on publicly available data and grounded in fair use," but the factual record in these cases keeps growing as courts grant broader discovery into AI training pipelines.

󠇟󠇠󠇡󠇢󠅱󠆕󠄎︈󠅼󠄠󠅲󠆎󠅈󠇁󠅸󠆒󠇒󠄐󠇥󠆧󠆄󠆕󠆋󠅠󠆢󠇀󠄨󠄳󠆚󠆙󠅍󠆲󠆳󠄧󠇙󠆤󠅙󠅎︅󠄬󠅃󠅁󠄌󠅇The Supreme Court's denial of certiorari in Thaler v. Perlmutter on March 2 left intact the lower court's holding that AI cannot be an author under the Copyright Act. 󠇟󠇠󠇡󠇢󠄫󠄦󠄡󠄲󠄆󠄧󠄗󠅲󠅤󠇑󠆜󠇫󠄁󠆬︈󠅽󠄛󠇙󠇛󠅛󠅠󠄈󠄡󠆄︉󠆒️︀󠅳󠄭󠇦󠄥󠅶󠆠󠆆󠅚󠄌󠆲󠄨󠅚That outcome leaves the authorship question where the district court placed it but does not resolve the larger training-data question. 󠇟󠇠󠇡󠇢󠅷󠄭󠄡󠅽󠇚󠇜󠄱󠇨󠇏︈󠆘󠅊󠅶󠄂󠅯︋󠅙󠄗󠄾󠄨󠆪󠅡󠅬󠄬󠆧󠆇󠄛󠆆󠅭󠇘󠆊󠅐󠄹󠆮󠆛󠅌󠆂󠄖󠄔󠆝Courts are resolving narrow issues while the commercial infrastructure question remains unanswered by litigation.

󠇟󠇠󠇡󠇢󠆻󠄔󠇟󠆖󠅭󠇌󠇭󠄶󠇃󠅁󠆲󠅙󠄾󠅞󠇜󠆰󠄦︁󠅘󠆆️󠅿󠆰󠅰󠆿󠆦︋󠅫󠄈︎󠇬󠄩󠆃󠇆󠇤󠄝󠇀󠆹󠇧󠄨Lawsuits create leverage. 󠇟󠇠󠇡󠇢󠅺󠆇󠇠󠇘󠆇󠆗󠄍󠅫󠆮󠇁󠇅󠆠󠇐󠄠󠇝󠆲󠄱󠄅󠅉󠄴󠇛︉󠇔󠆽󠆜󠅟󠅞󠆧︌󠄜󠆷󠅧󠄘󠅲󠆭󠆜󠅹󠆔󠆤󠅼The threat of statutory damages and the reputational cost of discovery make AI companies more willing to negotiate. 󠇟󠇠󠇡󠇢󠅙󠆥󠅷󠄿󠅦󠇪󠆙󠆞︌󠇄󠄢󠄩󠄌︃󠄉󠇬󠇎󠆪︍󠅚󠄗󠆝󠆛󠄌󠆽󠆸󠅴󠇧󠄫󠇠󠆒󠄷󠇩󠆍󠆯︄󠅞󠆜󠄽󠅰But leverage without infrastructure is like winning a judgment you cannot collect. 󠇟󠇠󠇡󠇢󠇟󠆆󠅈󠆟︈󠄩󠄊󠅾󠆚󠆸󠄂󠆾󠇜󠆕󠄮󠆪󠇃󠇍󠆦󠆯︊󠆟󠇦󠆣󠇄󠄪︊󠇄󠆺󠅔󠇥󠄛󠇤󠅰󠇄󠅀󠆊󠇄󠄙󠅘Even if courts rule definitively that training on copyrighted content requires a license, the publisher still needs a mechanism to identify which content was used, track that usage across AI systems, and enforce the terms of whatever license applies. 󠇟󠇠󠇡󠇢󠇩󠄇󠇩󠆸󠆱󠆖󠆌󠄀󠄥󠄵󠄷󠄱󠆦󠇡󠆶󠄓󠇁󠇏󠆝󠄻󠆗󠄝︄󠆙󠆓󠅔︇󠇀󠆶󠅴󠆣󠄃󠇗󠅗󠇁󠆌󠅬󠇅󠄍󠆔That mechanism is the same attribution infrastructure the marketplaces require.

󠇟󠇠󠇡󠇢󠅑󠇆󠅿󠄃󠆨󠄀󠅮󠄋󠇮󠇣󠅇󠆀󠅌︆󠅤󠅅󠄊󠇍󠄰󠆪󠄌󠅧󠅍󠆩󠄟󠄝󠆗󠄩󠆹︉󠆃󠅽󠄚󠇧󠄺󠇐󠆎󠇤󠇕󠅏The publishers who will benefit most from favorable court rulings are not the ones who filed the suits. 󠇟󠇠󠇡󠇢󠄐󠆎󠅽󠇃󠄋󠄊󠅟󠄵󠄙󠆀󠆥󠅋󠄄󠆶󠄻󠄇󠅂󠅘󠅍󠆍󠄪󠆳︄󠄠󠆨󠅨󠅓︄󠄉󠄭󠆻󠅘󠆫󠇕󠇇󠅣󠄏󠇆󠆻󠇭They are the ones who built the infrastructure to act on whatever rights those rulings establish.

󠇟󠇠󠇡󠇢󠄃󠇐󠄚󠄠󠆥󠇢󠄶󠇏󠅕󠇯󠆇󠅵󠅲󠄼󠆂󠇪󠆇󠄇󠆫󠇅󠇯󠇧󠅽󠄂󠄘󠅃󠄜󠄍󠅥󠇖󠄼󠄲󠅗󠄋󠄅󠇙︄󠆯󠇍󠅁The Regulatory Accelerant

The EU AI Act's Article 50 adds a hard deadline to this infrastructure question. 󠇟󠇠󠇡󠇢󠅄󠄟󠇆󠆟󠅯󠅷󠅸󠇋󠅮󠅂󠆣󠄵󠄱󠇌󠇯󠄯︂󠄯󠄬󠄧󠇘︉󠇝󠄘󠄧󠆈󠅐󠄡󠇐︊󠄀󠅨󠆄󠄧󠅶󠇧󠅶󠅤󠅆󠄦By August 2, 2026, providers of AI systems generating synthetic text, audio, image, or video must "ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated." 󠇟󠇠󠇡󠇢󠇖󠅕󠅹󠄭󠄅󠆋󠇆󠅟󠇣︌󠄎󠇮󠇀󠇠󠇙󠆴󠆸󠄓󠆃󠄥󠅗󠇨󠆽󠅃︀󠆫󠄴󠇮󠅁󠅏󠆳︌󠆪󠆎󠄜󠆝󠆦󠇄󠄔󠅭The European Commission published a draft Code of Practice in December 2025, with a final version expected in June 2026. 󠇟󠇠󠇡󠇢󠇫󠅐󠆿󠅹󠆞︅󠄲󠇈︁󠇦󠇏󠇏󠄆󠇇󠅣󠄽󠄞󠆼󠆜󠄕󠅷󠄖󠄍󠄪󠆤󠆢󠅳︌󠄩󠇔󠆆󠄲󠆅󠄯󠇢︃󠇄󠄽󠅦󠄭That Code of Practice is expected to reference C2PA-compatible standards for machine-readable marking. 󠇟󠇠󠇡󠇢󠄬󠆮︇󠅨󠄉󠅱󠇄󠅰󠅗󠄆︈󠅔󠆘󠇈󠆁󠅇󠄄󠇌󠇅󠄦󠆯󠄗󠆧󠇃󠅱󠄩󠅓󠇮󠆷󠅧󠅳󠇢󠅝󠆍︁󠇑󠇢󠅀󠇥󠇞Non-compliance carries fines of up to 35 million euros or 7% of global annual turnover.

󠇟󠇠󠇡󠇢󠅻󠄬󠇇󠅬󠄬󠄫󠆉󠄸󠆌󠇯󠇞󠅒󠄞󠅩󠄭󠆀󠆧󠅚󠄋󠅔󠇌󠄿󠅛󠇤󠄇󠅭󠄴󠄎󠅻󠄂󠅷󠆅︎󠅟󠆖󠇥󠅎󠇧󠄾󠄏Article 50 creates a regulatory backstop for the same infrastructure the marketplaces are already demanding. 󠇟󠇠󠇡󠇢󠆁󠆵󠇝󠇉󠇦︇󠆧󠄦󠇓󠅰󠇉󠄃󠅠󠅤󠄐󠅊󠇃󠅥󠆦󠆫󠇧󠄏󠆠󠅡󠆺󠄂️︈󠄯󠇛︅󠄎󠅎󠄌󠄝󠅶󠇕󠅓󠇑󠇋Publishers who build C2PA-compatible provenance systems get two things at once: access to licensing marketplace revenue and compliance with the EU's marking requirements. 󠇟󠇠󠇡󠇢󠅶󠇀󠅯󠆅󠆻󠆃󠆠󠄉󠇣󠆸󠆞󠅓󠅾󠆔󠇊󠆢󠅼󠆩󠆤󠅰󠆵󠆶󠇬󠇦󠄐󠆿󠇄󠅛󠄄󠆖󠅨󠇢󠆄󠄐󠇈󠇩󠄐󠆌󠇉󠅙Those who wait face a dual penalty - locked out of marketplace revenue and exposed to regulatory risk in their largest non-US market.

󠇟󠇠󠇡󠇢󠇖︃󠅹󠅩󠄀󠅃󠄍󠄞󠅷󠅩󠄅󠄕󠇭󠆩󠅀󠄐󠆄󠆽󠇊󠇋󠅐︉󠆸󠆨󠅺󠄣󠅘︎󠅫󠆦󠇉󠅨󠅕󠇀󠆘󠆘󠆚󠇯󠅰︍The August 2026 deadline also changes the calculus for AI companies. 󠇟󠇠󠇡󠇢󠅫󠆯󠄀󠅄󠄤󠇡󠆄︆󠆵󠆬󠄟󠅰󠇐󠅽󠆻󠇦󠆔󠄏󠇊󠄶󠅕󠄙󠆘󠆔󠆠󠆯󠇝󠄞󠆾󠇋︂󠄣󠇆󠄛󠄲󠇨󠅓󠅜󠇢󠅔Any enterprise deploying generative AI in the EU will need to demonstrate that its outputs carry machine-readable provenance markers. 󠇟󠇠󠇡󠇢󠅀󠇊󠆂󠄬󠆳󠅻󠇠󠄍󠇮󠆨󠄟󠇗󠄷󠇓󠆦󠆜󠇡󠆎󠅯󠄙󠆫󠄨󠅱󠅺󠄟󠅗󠇗󠇒󠄐󠇗󠅠󠅬󠆥󠅜󠆕󠄒󠅗󠄩󠆿󠄹That means the AI company needs to know the provenance of its inputs - which means it needs to source content from publishers whose material carries machine-readable signals. 󠇟󠇠󠇡󠇢󠇌󠇩󠆠󠄡󠅥󠄧󠅋󠄒󠇪󠆚󠄳󠄆󠄨󠅛󠆺󠅫󠇜󠄾󠄐󠇟︆󠇏󠄞󠇗󠅇󠅴󠇮󠆧󠄞󠄑󠆖󠅪󠅷󠅁󠇙󠇙󠅢󠅪󠄮󠆜The regulatory requirement strengthens the commercial case for provenance infrastructure on both sides of the market.

󠇟󠇠󠇡󠇢󠄙󠆰󠇬󠆈󠇇󠅯󠇟󠅳󠄥󠆸󠆂󠅮󠆡󠅬󠄠󠆁󠅻󠅡󠇧︂󠆮󠇄󠅂󠅆󠄏󠆀󠅠󠅪󠄸󠆝󠇂󠅲󠇐󠇇󠆭󠄽󠆢󠄒󠇓󠇁Where the Revenue Goes Next

AI companies are building licensing marketplaces because their enterprise customers demand provenance assurance. 󠇟󠇠󠇡󠇢󠅺󠄽︂󠆦󠄋󠆇︀󠄁󠆝󠅬󠇋󠅆︃󠅕󠆡󠅀󠅘󠆣󠆓󠅚󠄟󠅱󠅟󠄥󠅣󠇅󠇌󠆐︎󠆜󠅊󠅓󠇜󠇩︃󠄾󠄌󠆃󠇈󠆍Regulators are mandating machine-readable markers on AI outputs, which forces AI companies to track provenance on inputs. 󠇟󠇠󠇡󠇢󠄶󠇮󠄭󠅝󠄪󠄽󠅊󠇛󠇮︇︊󠆥󠅬󠄗󠄉󠅜󠄇󠅭󠄭󠇌︋󠅶󠆈󠄲󠆏󠄂󠅬󠇏󠆦󠄷󠆞󠄻󠆁󠇏󠅚󠄻󠄤󠄨󠄢󠇡Publishers who embed provenance signals into their content become discoverable, attributable, and payable in these systems. 󠇟󠇠󠇡󠇢󠇍󠆆󠅷󠅲󠄾󠆗󠆽󠇠󠄟󠆝󠇙󠇮󠄲󠅁󠅯󠄙󠇙󠇠󠄌󠆔︀󠇖󠇠󠅕󠄜󠇘󠄹󠆥󠄵󠅤󠅣󠄢󠆼󠅆󠆐󠄷󠆪󠅐󠅬︋Publishers who do not remain invisible.

󠇟󠇠󠇡󠇢󠇋︄󠅎󠄥󠆨󠄲󠆃󠆥󠆅󠅁󠄅󠆤󠄤󠆸󠆹󠄴󠄀󠅤󠆚󠅠︃󠆁󠅠󠆼󠅠󠆠󠇔󠆵󠆹󠄱󠆈󠆩󠇎󠅌󠆝󠄕󠆄󠇟︃󠇆Litigation will continue and will produce important precedent. 󠇟󠇠󠇡󠇢󠆳󠇜󠆨󠄋󠄘󠆸󠆸󠅝󠇏󠄄󠄼󠆽󠅆󠄛󠄰︁︉󠇧󠆀󠅨󠄺󠇞󠄣󠇊󠆁󠆣︎󠇋󠅬󠆢󠆳󠄠󠅻󠇃󠅆󠅈󠄯󠆪󠆈󠅕The Britannica case, the training-data disclosure orders, and whatever the Supreme Court eventually says about fair use in the AI context all matter. 󠇟󠇠󠇡󠇢󠆫󠅿️󠅝󠇣󠇣󠅴󠆠󠆹󠇇󠄳󠆣️󠅔󠇊󠅤󠇏󠆂󠆳󠆿︉󠄖󠇠󠇢󠄪󠄳󠇔󠅾󠇦󠅥󠇃󠄷󠇨󠄛󠆣󠄸󠇉󠅝󠆵󠅨But precedent without infrastructure produces legal clarity and commercial paralysis. 󠇟󠇠󠇡󠇢󠄵󠅤󠆶󠄜󠄖󠆂󠆔󠄀󠄹󠄓󠇨󠄄󠇣󠅰󠆣󠅧󠆡󠆄️󠆀󠇩󠅍󠅆󠄑󠄟󠆜󠅉󠆐󠄬󠇨󠆔󠄑󠆞󠆧󠇄󠄆󠄠󠇃︍󠅽The publisher who wins the right to be paid but cannot be identified by the system writing the check has won a pyrrhic victory.

󠇟󠇠󠇡󠇢󠄌󠅻󠆄󠅯󠄗󠄩󠆋󠄝󠆟󠆺󠇁󠄔󠅸󠆇󠆟󠄉︉󠄞󠆔󠆂󠅲󠄀󠆵󠅡󠅀󠅪󠆬󠅝︈󠅵󠄎󠇊󠆜󠆇󠆒󠆗󠆒󠇚󠄚󠄤Within twelve months, the publishers capturing meaningful AI licensing revenue will not be those who filed the most lawsuits. 󠇟󠇠󠇡󠇢󠆊󠇦󠇯󠇐󠆜󠄼󠄗󠅳󠇢󠅕󠇠󠅡󠆆󠄔󠄌󠅪󠄾󠇔󠄾󠅘󠄈︃󠅭󠄶󠅕󠆗󠄅󠅳󠅑󠆞󠇌󠄺󠇒󠆨󠅽󠆋︄󠄇󠆦󠆭They will be those who made their content machine-readable first. 󠇟󠇠󠇡󠇢󠄍󠅻󠄈󠄅󠇧󠅠︂󠆹󠇮󠇜󠅻󠄌󠆯󠄿󠆫󠆻󠆟󠄡󠆩󠇠󠄧󠄮󠇬󠅩󠅴󠅨︁󠄜󠄞︈󠇁︀︂󠅅󠄓󠅗️︈󠆁󠄚The NMA/Bria deal and Microsoft's PCM are the early evidence. 󠇟󠇠󠇡󠇢󠆪󠅦󠅦󠄠󠄆󠇐󠅲󠆒󠅕󠅞󠆘󠅤󠄤󠆚󠆼󠅢󠄖󠆗󠅤󠇥󠇮󠅱󠄩󠅏︀󠆟󠆤󠇜󠇬󠆆󠇬󠆪󠆚󠄗󠄇󠆞︇󠅵󠅶󠅉The EU AI Act's August deadline is the forcing function. 󠇟󠇠󠇡󠇢󠆷󠅸󠇍󠇝󠄶󠆩󠅲󠄢󠆵󠄁󠄁󠄺󠇚󠅔󠆗󠄴󠄉󠇔󠆌󠆯󠄡󠇨󠅛󠄬󠄄︉󠆪󠇥󠄍󠅓󠅃󠄃󠅊󠄠󠆇󠇉󠄔󠄧󠇒󠄛The first major publisher to report meaningful recurring AI licensing revenue will have invested in provenance infrastructure before it invested in litigation. 󠇟󠇠󠇡󠇢󠆠󠄏󠅧󠅵󠄊󠅥󠇔󠅦󠆽󠇐󠆪︃󠅽󠅬󠅳󠆂󠆪󠇃󠆓󠇅󠅮󠆭󠄚󠇬󠆩󠄲󠆳󠇌󠅋󠆦󠄪󠅩󠆮󠄏󠆁󠇡󠆪󠅍󠄯󠅽That publisher is probably already in the Microsoft PCM beta. 󠇟󠇠󠇡󠇢︃󠇋󠆢󠅍󠄧󠆢󠄸󠆎󠅈󠅞󠅪󠇖󠆧󠅈󠅓󠅔󠆌󠇨󠆕󠅉󠆏󠅼󠇠󠅪󠄈󠄍󠅑󠅡󠅤󠅁󠇫󠅰󠇜󠆃󠇘󠆭󠄔󠅌󠆃󠆝