Google defends AI training as fair use amid publisher pushback

Image credit: Search Engine Journal

Google published a policy paper Tuesday in the United States, asserting that training artificial intelligence models on publicly available web data constitutes fair use under U.S. copyright law, amidst growing demands from publishers and regulators for attribution and compensation.

The technology giant’s stance, detailed in a paper titled “A Pragmatic Approach to AI Governance in America,” directly challenges calls for a permission-first approach to content scraping for AI development.

Google argued that using public web data for AI model training is a “transformative, non-expressive use” akin to an art student drawing inspiration from existing works, according to the document.

The company suggested that existing copyright law, alongside machine-readable controls such as Google-Extended in robots.txt files, should serve as primary mechanisms for publishers to manage their content.

The paper also noted that Google is exploring partnerships and paid agreements for specialized content and has established notice-and-takedown processes for AI outputs that directly copy existing work.

However, publishers and regulatory bodies have pushed back against Google’s position, seeking more protections beyond simple opt-out mechanisms.

Digital Content Next, an organization representing premium publishers, sent a cease and desist letter to the Common Crawl Foundation, stating that “copyright law is not an opt-out regime,” directly refuting Google’s argument.

Regulators, including the United Kingdom’s Competition and Markets Authority (CMA), have also called for clearer attribution, compensation, and permission-based scraping practices for AI training data.

Google’s policy positions outlined in the paper are not firm product commitments, and the specifics regarding potential value-exchange programs or licensing terms remain flexible, the company indicated.

Source: Search Engine Journal