US backs OpenAI, calling LLM training on news fair use

This is one of the most consequential legal signals yet for the economics of frontier AI. By explicitly arguing that training LLMs on copyrighted works is fair use, the US government is telling courts that aggressive licensing requirements would threaten national competitiveness and even national security. That is a powerful framing for judges who may be wary of kneecapping a strategic technology industry.([techcrunch.com](https://techcrunch.com/2026/09/02/u-s-government-sides-with-openai-on-issue-of-training-llms-on-copyrighted-material/))

If courts lean toward this position, it massively reduces legal overhang for labs like OpenAI, Anthropic and Google, which have already trained on vast corpora of unlicensed text. It also weakens the bargaining power of large rights‑holders who hoped to extract recurring training fees, tilting the balance toward one‑time settlements or output‑focused remedies instead. That in turn favors well‑capitalized labs that can afford the initial legal fights and infrastructure, and disadvantages smaller players who might have relied on licensed niche datasets as a differentiator.

From an AGI‑timeline perspective, a strong fair‑use precedent speeds things up. It keeps the cost and friction of assembling training data closer to the status quo, encourages continued scaling of multimodal corpora, and reduces the odds that a court injunction slows or pauses major training runs. The risk, of course, is that weak copyright protections may leave creators with fewer tools to shape how their work is used in ever‑more capable systems.