Google's Fair Use Push: AI Regulation Splits Courts and States

Two federal judges in San Francisco looked at the same question – whether training AI on copyrighted web data qualifies as fair use – and came to opposite conclusions within days. One called the practice "quintessentially transformative." The other warned it could destroy the economic incentives that make creative work possible.

The split was not a surprise. It is the natural result of a regulatory environment that has also gone in different directions. California and the European Union now require AI companies to disclose training data sources. Colorado revised its AI Act in May to focus on real-world outcomes instead. Google, in a 21-page governance paper published Thursday, argued that the second approach is the right one.

"This approach would address outputs, not inputs," the paper stated. "Looking to prevent and mitigate specific harms rather than micromanaging the science behind these new tools." The paper is Google's most direct statement yet on how AI regulation should work. It aligns with Colorado's revised framework, which requires companies to determine where AI materially influences a consequential decision and notify consumers if that leads to an adverse outcome.

On copyright, Google argued that training on publicly available data is a "transformative, non-expressive use – like an art student taking inspiration from walking through a gallery – that should remain protected under fair use in the U.S." The company pointed to machine-readable opt-out controls like robots.txt as the appropriate mechanism for publishers who want to block their content.

Publishers are not buying that argument. Digital Content Next sent a cease-and-desist letter to the Common Crawl Foundation, arguing that copyright law "is not an opt-out regime." That directly challenges Google's position.

The judicial record mirrors the divide. U.S. District Judge William Alsup in San Francisco ruled that AI training is "quintessentially transformative" and said copyright law "seeks to advance original works of authorship, not to protect authors against competition." Two days later, U.S. District Judge Vince Chhabria, also in San Francisco, warned that widespread AI training could undermine the economic incentives that drive human creative work. More cases involving Anthropic, Google and Stability AI are pending in 2026.

Daryl Lim, H. Laddie Montague Jr. Chair in Law at Penn State Dickinson Law School, told PYMNTS that only a handful of firms can train frontier models at scale because those firms simultaneously control compute, data, cloud infrastructure, and distribution. "When you train frontier models, you need to ingest vast repositories of works that may include copyrighted works," Lim said. That concentration is the structural concern that an output-based framework alone does not address.

For investors, the key risk is not the next court ruling in isolation. It is the absence of a consistent legal framework across jurisdictions. Companies like Google, OpenAI, and Anthropic face different rules depending on where they operate and which judge hears the case. The Colorado approach provides one path forward. The California and EU approach provides another. The courts are still deciding which one wins.

The Google paper makes a clear bet on the output-first model. The next set of rulings, due in 2026, will test whether that bet holds up in court.