The Landscape Changed in 2024-2026
Since generative AI spread in 2023, training-data lawsuits surged worldwide. The central issue: "is unauthorized training fair use or copyright infringement?" Case law is accumulating, with conclusions diverging by region. Here are the main cases.
Text
NYT v. OpenAI / Microsoft (filed 2023.12)
The New York Times presented 100+ examples of "ChatGPT verbatim-reproducing NYT articles." It argues both unauthorized training-data use and brand dilution. A settlement or ruling in 2025-2026 is expected, an important suit that will be the industry's benchmark.
Others
- Authors Guild v. OpenAI (unauthorized training on books)
- Newspapers like Wall Street Journal, CNN, Daily Mail filing one after another
- OpenAI also concludes license contracts with AP, Axel Springer in parallel
Image
Getty Images v. Stability AI (US filed 2023.2, UK filed 2023.1)
Stock-photo giant Getty argues "Stable Diffusion trained on Getty's watermarked images." The UK ruling (2025) is partly favorable to Getty; the US continues into 2026. It includes the trademark-dilution issue, making it composite.
Artist Class Actions
- Sarah Andersen et al. v. Stability AI / Midjourney / DeviantArt
- Class action by Karla Ortiz et al. (US California)
- Issues: copyright scope of style imitation, legality of style transfer
In June 2026, Getty Images struck a deal with OpenAI to surface licensed Getty imagery inside ChatGPT responses (Innovatopia). While Getty continued its copyright case against Stability AI, it chose a licensing partnership rather than litigation with OpenAI — signaling the image-AI debate is shifting from "is training legal?" lawsuits toward commercial deals that bring rights-cleared stock into AI products. ChatGPT gains cleaner provenance for visual outputs, and Getty unlocks a new AI-driven licensing revenue channel.
Music
RIAA v. Suno / Udio (filed 2024.6 → 2025 settlement reported)
The US recording association and Sony/Universal/Warner Music class-sued Suno and Udio. Argues "can reproduce songs by ABBA or Mariah Carey." Multiple media reported it reached a settlement + license contract in 2025. Though terms are undisclosed, it became a turning point spreading a "data provision + revenue sharing" model in the industry.
Code
GitHub Copilot Lawsuit (filed 2022.11 → 2024 dismissed)
The class action over OSS-license compliance had main claims dismissed in 2024. Verbatim code reproduction is still an issue, but it was judged that proving the illegality of "the training act itself" abstractly is a high bar.
Japan's Situation
- Protest movements and litigation warnings against AI voice-changers that trained on voice actors' voices without permission
- Illustrator training exclusion (NoAI tag, progress from the Mimic controversy)
- The ACA's "thinking on AI and copyright" (March 2024) is a practical guideline
- Individual licensing: Kodansha, Shogakukan etc. started AI-training-data licensing businesses
Organizing the Issues
| Issue | Dispute status |
|---|---|
| Infringement of the training act | US fair use vs individual judgment, UK negative |
| Verbatim reproduction (memorization) | Toward infringement findings with direct evidence like NYT v. OpenAI |
| Style protection | Out of copyright in most jurisdictions; limited protection possible under unfair-competition law |
| Data ownership/contract terms | License deals becoming common (NYT, AP, labels) |
What Companies Should Do Now
- Confirm the training-data provenance of the AI model used (inquire the vendor)
- Raise the ratio of commercial-license and opt-in material (Adobe Firefly, Getty AI, etc.)
- Self-check whether output resembles existing works
- State an AI-use policy in creator contracts
- Retain usage logs ~5 years for future law changes
New Type: Lawyer Liability from AI Hallucination
In 2025, in the US lawsuit Concord Music Group v. Anthropic, court filings by the major law firm Latham & Watkins were found to contain hallucinations (fictional/incorrect info) generated by Claude. The lawyer-ethics/disciplinary risk when AI output is submitted without sufficient human verification is being questioned, raising a new type of litigation/sanction risk—"professional liability for AI-generated misinformation"—on a separate axis from copyright disputes.
Summary
The relationship of generative AI and copyright is in a transition where "black or white" is decided by both litigation and contracts. Until case law settles, self-defense via license deals and internal guidelines is wise. Continuously watch NYT v. OpenAI and Getty v. Stability's final conclusions as important suits directly tied to industry-wide rule formation.
