Back to News
PolicyAI Understanding briefing

US appeals court upholds Thomson Reuters copyright win over ROSS AI search engine

A U.S. Court of Appeals affirmed Thomson Reuters' copyright claim against ROSS Intelligence, rejecting the legal‑research startup’s fair‑use defense for using Westlaw headnotes to train its AI‑driven legal search tool.

4 min readRead the linked source
Source-provided image accompanying US appeals court upholds Thomson Reuters copyright win over ROSS AI search engine
Source referenceSource recorded
Publisher
cryptopolitan.com
Source link
cryptopolitan.comhttps://www.cryptopolitan.com/appeals-court-upholds-thomson-reuters-ai-training-copyright-win-over-ross/
Source type
Linked source — primary-source status has not been established.
ContextUnderstand this in 60 seconds

Start here

Key terms

Generative AI
AI systems that produce new content such as text, images, audio, video, or code.
Test yourselfAI Ethics Quiz

What happened

The 2nd U.S. Circuit Court of Appeals affirmed a lower‑court ruling that ROSS Intelligence infringed Thomson Reuters’ copyrights by copying 2,243 Westlaw headnotes to develop its AI‑powered legal research platform, rejecting ROSS’s fair‑use argument.

On September 29, 2026, the U.S. Court of Appeals for the 2nd Circuit upheld a decision originally issued by Judge Stephanos Bibas in 2023. The appellate panel affirmed that ROSS Intelligence directly infringed Thomson Reuters’ copyrights by copying 2,243 Westlaw headnotes, which it used to build a competing legal‑research search engine.

ROSS had previously sought a licensing agreement for Westlaw content but was denied because it competes with Thomson Reuters. Instead, ROSS obtained roughly 25,000 “Bulk Memos” from LegalEase, which were derived from Westlaw headnotes. The court found that ROSS’s use was not transformative, as it merely repackaged the headnotes to facilitate its own search tool, rather than creating new expressive content.

Judge Bibas, whose earlier ruling the appeals court affirmed, emphasized that the case involved “only non‑,” distinguishing it from generative large language models. The court’s analysis focused on the amount copied, the purpose of the copying, and the effect on the market for the original works, concluding that fair use did not apply.

Source details: cryptopolitan.com ↗

Why it matters

The decision clarifies that non‑ tools that reproduce copyrighted legal content may not qualify for fair use, signaling that AI developers could face higher licensing costs for proprietary data. The ruling could reinforce the need for formal licensing agreements and may influence how AI firms source training material, potentially widening the gap between large incumbents and smaller innovators.

The ruling provides a concrete legal precedent that AI‑driven tools which reproduce copyrighted material without substantial transformation may be liable for infringement. This could compel AI firms to negotiate costly licensing deals for high‑quality proprietary datasets, reinforcing the advantage of well‑capitalized companies that can afford such fees.

The decision also signals to the U.S. Copyright Office and policymakers that existing fair‑use doctrine may be insufficient for the scale of modern AI training. The Office has already noted that fair‑use determinations are fact‑specific, especially for , suggesting that future cases could yield different outcomes depending on the technology used.

Industry analysts, including the OECD and Goldman Sachs, have warned that data access, computing power, and licensing costs are key competitive levers in AI. This case underscores the risk that tighter control over data could concentrate AI capabilities among a few large firms, potentially stifling innovation from smaller entrants.

Interactive Mechanism

Interactive Mechanism: How It Actually Works

Explore the underlying technology behind this development interactively.

Thinking Budget (Test-Time Tokens):1,024 tokens
Complex Accuracy79%Math & Code Logic
Latency3.2sTime to first full output
Inference Cost$0.0092Per query estimated
Reasoning StyleStep VerificationInternal chain depth
Active Thinking Trace:
1Deconstruct user problem into formal constraints
2Propose candidate hypotheses & step-by-step calculation
3Self-correction: Backtrack and refute subtle edge cases
4Exhaustive consistency check & final output synthesis
Core takeaway: Test-time compute fundamentally changes AI economics. Instead of only scaling during pre-training, giving reasoning models more tokens at inference time allows them to systematically solve PhD-level STEM problems.
Interactive Concept Check+10 Points
AI Ethics Quiz

Impossibility results in algorithmic fairness (e.g. Kleinberg et al., Chouldechova) show what?

What to watch next

Future litigation will test whether the reasoning extends to generative large language models, and lawmakers may consider statutory reforms or compulsory licensing schemes to address AI‑training data at scale.

Watch for appellate or Supreme Court challenges that may broaden or narrow the scope of fair use for AI training, particularly as generative models become more prevalent.

Legislative activity in the U.S. and Europe may introduce compulsory licensing frameworks or voluntary licensing markets to address the data‑cost barrier highlighted by the ruling.

Monitor how other AI companies respond—whether they pursue new licensing agreements, develop alternative data‑collection strategies, or adjust model architectures to reduce reliance on copyrighted content.

Related guides & quizzes

AI EthicsAI TrainingFuture of AITest what you know — try a free AI quizLook up an AI term in our glossaryFollow the AI regulation tracker
Found this useful?