AI Training Fair Use Status

was this overturned or is it still true? https://www.dglaw.com/court-rules-ai-training-on-copyrighted-works-is-not-fair-use-what-it-means-for-generative-ai/

Here’s the **current legal landscape (as of late-2025)** regarding the claim in the article you linked about *AI training on copyrighted works not being fair use*: navlistLatest on Fair Use & AI Training (2025)turn0news1,turn0news4,turn0news0 ### ✅ **1. That older ruling (Ross Intelligence) still stands — but it’s narrow** The case cited in the DGLaw article (*Thomson Reuters v. Ross Intelligence*, Feb **2024**) held that **using copyrighted works to train a specific AI system was *not* fair use** — but: - It involved a **non-generative AI legal search tool**, not a typical generative LLM like ChatGPT/Claude. - The court emphasized the use was *not transformative* and competed directly with the copyright owner’s market. citeturn0search10 So the **general statement in that article — that training AI on copyrighted material “is not fair use” — is outdated and too broad**. It’s true *for that particular case*, but not a universal rule for all AI or generative models. ### ✅ **2. More recent court decisions actually *favor* fair use** In mid-2025, courts in the **Northern District of California** issued rulings saying: - In *Bartz v. Anthropic*, training generative AI models on copyrighted works can be **fair use**, especially when the use is highly transformative and doesn’t meaningfully harm the market for the original works. citeturn0search11turn0search29 - In *Kadrey v. Meta*, a similar result was reached — again focusing on the specific facts of the case (e.g., evidence of market harm). citeturn0search11 These rulings mark some of the **first major government decisions recognizing that training generative AI can qualify as fair use**, although they are still fact-specific. ### ✅ **3. Courts emphasize *fact-specific analysis*** Across multiple opinions and guidance from the U.S. Copyright Office: - Fair use is *not* a blanket rule for all AI training — it depends on how the data was used, whether the use is transformative, and whether there’s market harm. citeturn0search12 - Judges explicitly note that different facts could lead to different outcomes. This means: - **Some training on copyrighted works *is* fair use** - **Some isn’t fair use** - The legal line isn’t fully settled yet ### 📌 **Bottom line** ✔ The DGLaw article accurately described one court’s ruling, but ✘ **it is *not* the final word on AI training and fair use.** As of late-2025: - **Generative AI training *can be* fair use** under U.S. copyright law in some cases (recent rulings support this). citeturn0search13 - The issue remains *uncertain and fact-specific*. Courts continue to weigh factors like transformativness and market harm. citeturn0search11 If you want, I can summarize the key differences between the major AI copyright cases or explain what these rulings mean for developers and creators.