Nikolin Muçaj has posted Scrape, Train, Compete: Data Extraction as Unfair Competition in the AI Era on SSRN. Here is the abstract:
Artificial intelligence companies increasingly scrape vast quantities of online content to train systems that substitute for, rather than refer users to, their sources. When tools such as ChatGPT generate responses derived from news reporting by outlets like The New York Times, they compete for the same reader attention that publishers cultivated through sustained investment in journalism. Yet existing legal frameworks provide little recourse. Copyright law excludes most factual and informational works under the originality requirement articulated in Feist, while common-law misappropriation doctrine has been narrowed almost to extinction by federal pre-emption.
This paper addresses that gap by developing a market substitution framework for AI-era scraping disputes. Liability arises where content appropriation (i) extracts material from substantial investment, (ii) enables competing products, and (iii) would systematically undermine creation incentives. The framework survives pre-emption by targeting competitive displacement, an element distinct from copyright’s reproduction right, while requiring actual market harm rather than mere copying.
Neither copyright nor existing misappropriation tests adequately address substitutive scraping. Copyright’s originality threshold forecloses protection for factual compilations, while the “hot news” doctrine articulated in NBA v. Motorola, with its requirement of time-sensitive, near-simultaneous reuse, is ill-suited to large-scale training data extraction. Contractual and technological controls likewise fail at web scale due to coordination and enforcement limits.
Market substitution analysis avoids these limitations and survives pre-emption scrutiny. It targets competitive harm, an element absent from copyright’s exclusive rights, while respecting Feist’s rejection of “sweat of the brow” protection by requiring proof of actual market displacement. The framework also draws a distinction between complementary uses that expand markets, such as search engine indexing, and substitutive uses that cannibalize demand, including AI-generated news summaries and synthetic content that replaces subscriptions.
Applied to The New York Times v. OpenAI, the analysis shows that training LLMs on news archives to generate competing content constitutes paradigmatic substitution. ChatGPT competes directly for reader attention, and, if widely adopted, would severely undermine incentives for news production. Without legal accountability for substitutive uses, AI developers capture content value while original producers bear creation costs, risking a tragedy of the commons in information markets.