The dispute turns on whether training generative AI on copyrighted works qualifies as fair use, and the U.S. Copyright Office says there is no single answerCITE:E1. The agency's May 2025 report frames transformativeness as a matter of degree that depends on a model's functionality and how it is deployed, not a binary testCITE:E1. The Office's own language captures the ambiguity: "transformativeness is a matter of degree, and how transformative or justified a use is will depend on the functionality of the model and how it is deployed"CITE:E1.
Two Anthropic cases produced opposite mechanisms for resolving the same underlying dispute: a court ruling and a negotiated settlement. In June 2025, Judge William Alsup ruled that training LLMs on lawfully acquired books is "exceedingly transformative" and constitutes fair use, but held that building a central library from pirated sources is not protectedCITE:E3. Alsup was explicit about the piracy distinction: "The person who copies the textbook from a pirate site has infringed already, full stop"CITE:E3. That piracy exposure translated into a monetary outcome in the related Bartz v. Anthropic case, which ended in settlement rather than a damages verdict: Anthropic agreed to pay a minimum of $1.5 billion, covering works drawn from pirated book repositories, at roughly $3,000 per class workCITE:E2. The court granted final approval of the settlement on July 20, 2026CITE:E2.
The New York Times' copyright suit against OpenAI and Microsoft remains in the discovery and pretrial phase as of 2026, with no ruling on the meritsCITE:E5. The Times alleges its articles were copied without authorization to train AI models, a claim that has not been adjudicatedCITE:E5. On January 5, 2026, the court issued an order in the multidistrict litigation compelling production of roughly 20 million sampled ChatGPT conversation logs — a discovery dispute, not a fair-use findingCITE:E5.
Yes: a Delaware federal court ruled in February 2025 that Ross Intelligence's unauthorized use of Thomson Reuters' Westlaw headnotes to train a legal search tool was not fair use, distinguishing it from generative AICITE:E4. The court reasoned that Ross's AI "is not generative AI (AI that writes new content itself)"CITE:E4, treating the absence of new-content generation as a key factor against fair use.
Content owners are pursuing two tracks simultaneously: asserting that unauthorized use is inherently unfair, and signing paid licensing deals with AI developers. The Authors Guild argues it is "inherently unfair to use and incorporate books, articles, and other copyrighted works in the fabric of AI technologies without the author's consent, compensation, or credit"CITE:E6, and calls for AI training to require authorization and paymentCITE:E6. In parallel, AI developers have moved toward direct licensing: OpenAI signed a deal with Axel Springer on December 13, 2023, to pay for use of its news content in model training and in ChatGPT, with the payment amount undisclosedCITE:E7.
| Case | Entity | Date | Outcome |
|---|---|---|---|
| Bartz v. Anthropic | Anthropic | 2026-07-20 | Settlement: ≥$1.5B, ~$3,000/work, ~500,000 worksCITE:E2 |
| Training-on-books ruling | Anthropic | 2025-06-23 | Court ruling: training fair use; pirated library notCITE:E3 |
| Thomson Reuters v. Ross Intelligence | Thomson Reuters | 2025-02-11 | Court ruling: non-generative AI use not fair useCITE:E4 |
| NYT v. OpenAI/Microsoft | The New York Times | 2026-01-05 | Ongoing: discovery order for ~20M chat logsCITE:E5 |
| Axel Springer licensing deal | OpenAI | 2023-12-13 | Licensing agreement, amount undisclosedCITE:E7 |
Taken together, the evidence shows a legal landscape splitting along two axes rather than converging on one rule: lawful-source training versus pirated-source training, and generative versus non-generative AI. Alsup's ruling and the Ross Intelligence decisionCITE:E3CITE:E4 point toward fair-use protection narrowing around how content was acquired and what the AI produces, while the Bartz settlement's per-work payoutCITE:E2 and the Axel Springer dealCITE:E7 show the same underlying dispute now also being resolved commercially, outside the courtroom, even as the Authors Guild's consent-and-compensation demandCITE:E6 and the still-unresolved NYT litigationCITE:E5 indicate the legal question is far from settled.
The dispute turns on whether training generative AI on copyrighted works qualifies as fair use, and the U.S.
Two Anthropic cases produced opposite mechanisms for resolving the same underlying dispute: a court ruling and a negotiated settlement.
The New York Times' copyright suit against OpenAI and Microsoft remains in the discovery and pretrial phase as of 2026, with no ruling on the meritsCITE:E5.
Yes: a Delaware federal court ruled in February 2025 that Ross Intelligence's unauthorized use of Thomson Reuters' Westlaw headnotes to train a legal search too…
The two Anthropic outcomes matter more as a pair than individually: a court found training on lawfully acquired books fair use, yet the same company still paid out for the pirated portion of its dataset at roughly $3,000 per work across close to 500,000 works. That gap shows fair-use protection is increasingly a function of sourcing, not just of what a model does with the data. The Ross Intelligence ruling adds a second axis — generative versus non-generative — that non-generative AI tools can't rely on. With that split now visible, the metric worth tracking next is how the NYT litigation resolves once discovery on the 20 million sampled ChatGPT logs is complete, since a merits ruling there would test whether Alsup's transformativeness reasoning holds for a plaintiff alleging direct verbatim output rather than a pirated training library.
Utility-scale solar has become one of the cheapest new power sources, with costs down roughly 90% since 2010, while 2024 deployment volume outpaced every other generation technology. Paired with record-low battery prices, solar-plus-storage is already powering an AI data center in Arizona — though it still falls short of full 24-hour dispatchable baseload.
2026 marks the European Union AI Act's real enforcement start: the AI Office gains fining power over general-purpose AI on August 2, while high-risk system deadlines are pushed to 2027 and 2028. California activates two new laws on January 1 covering frontier-developer safety disclosure and training-data transparency. The federal government moves the opposite direction, ordering a Justice Department task force to challenge state AI laws, naming California's SB 53 as a target.
China's EHang (億航) is the only eVTOL maker actually flying paying passengers today, holding a full Chinese type, production, and operating certificate set and delivering 221 aircraft in 2025. US rivals Joby and Archer remain in certification or pre-launch stages, while Germany's Lilium and Volocopter both went insolvent in 2024–2025, with Volocopter absorbed by a Chinese buyer.