1. United States Federal Precedent: Key Cases
Bartz v. Anthropic (Northern District of California, 2025)
Judge Alsup granted partial summary judgment in favor of Anthropic regarding the use of legitimately purchased books scanned and digitized for training its LLM, Claude.
The court found such use to be “spectacularly transformative,” akin to human learning—thus qualifying as fair use. It held minimal market harm and no reduction in demand for the original works .
However, the court rejected fair use for materials obtained by piracy, holding that pirated works—even for transformative AI training—were inherently infringing .
Those claims (relating to pirated inputs) are to proceed to trial .
Meta (Kadrey) v. Meta Platforms (Southern District, Judge Chhabria, 2025)
Authors sued Meta, alleging that training its AI on their works without permission infringed copyright.
Judge Chhabria dismissed the case, concluding that plaintiffs failed to present sufficient evidence of market harm and inadequate argument on copyright violation, while noting Meta's use was transformative under fair use doctrine .
However, the decision was narrow—limited to the 13 plaintiffs—and does not establish general legality for AI training practices involving copyrighted material .
Thomson Reuters v. Ross Intelligence (District of Delaware, 2025)
Ross used Westlaw headnotes to train its legal‑research AI.
The court granted partial summary judgment for Thomson Reuters on infringement and denied the fair use defense. Ross’s use was found not transformative, and it created direct market competition, thus failing the fair use test .
Significance: first on‑the‑merits rejection of fair use for AI training—even though this was non‑generative AI .
2. Image‑AI Litigation: Art in the Courts
Andersen, McKernan & Ortiz v. Stability AI, Midjourney, DevianArt (N.D. California)
Artists allege infringement from billions of images scraped without consent.
In early U.S. ruling the court declined to dismiss the complaint entirely, allowing amended claims—specifically, allegations that plaintiff works were used in the LAION dataset remain viable .
Getty Images v. Stability AI (UK High Court)
Getty sued Stability AI in the UK alleging unauthorized use of Getty images to train Stable Diffusion.
The High Court refused to strike the claims, and the dispute over whether their software qualifies as an “article” under UK copyright law is set for trial in summer 2025 .
3. Other Jurisdictions & Context
Italy
Supreme Court of Cassation confirmed that a graphic work created via AI can qualify as creative and copyright‑protected if the human user's input substantially shaped the output—requiring factual assessment of the human’s creative role .
Philippines
The Copyright Bureau recently registered an AI‑generated poster, holding that a user's prompt input and creative editing may satisfy the legal requirement of human authorship under the Intellectual Property Code. It recognized AI‑assisted works as eligible—even absent classical human authorship—placing them in a distinct category under Class O .
IPOPHL has formal guidelines (2024) clarifying fair‑use limitations—though still no Philippine cases deciding on AI‑training infringement or outputs .
4. General Observations & Trends
U.S. courts have rendered mixed rulings: legitimate acquisition + transformative use → fair use (Anthropic, Meta), but pirated inputs or competitive replication → infringement (Ross Intelligence).
Plaintiffs must emphasise market harm and evidence of direct competition to challenge AI training practices effectively .
The question of whether AI‑generated outputs infringe on protected works remains largely unresolved—most rulings address the legality of ingestion (training), not the outputs themselves.
Legislative developments (e.g., proposed U.S. bills) could reshape the terrain, introducing stricter liability and licensing requirements for AI training on copyrighted content .
The law is still unsettled; outcomes vary case‑by‑case, jurisdiction‑by‑jurisdiction, and depend on factual specifics (type of data, acquisition method, transformation level, market impact) .
🧾 Summary Table
Jurisdiction Use Scenario Court Ruling Fair Use Outcome
U.S. – Anthropic Scanned books lawfully purchased Transformative use; no market harm Fair Use accepted
U.S. – Anthropic Books from pirated sources Piracy is inherently infringing No Fair Use
U.S. – Meta Training on authors’ books Plaintiffs failed, lacked market-harm proof Fair Use accepted
U.S. – Ross Int’l Copying legal headnotes for AI Commercial use; competing product; not trans. Fair Use rejected
U.S. – Artists Scraped images for image‑AI training Complaint survives; further litigation needed Undecided
UK – Getty Images Image ingestion into Stable Diffusion Claims proceed to trial Pending
Italy AI image created with user input Eligible for copyright if human contribution Recognized
Philippines AI artwork via user prompt + editing Human creator qualifies under IP Code Registration accepted
⚖️ In conclusion, courts continue to recognise fair use for transformative training uses based on legally obtained material. Yet, cases fail where inputs are pirated or the AI tool competes directly with the original market. Outputs themselves remain an emergent battleground. Legal outcomes will hinge on fact‑intensive analyses of acquisition, transformation, and economic impact.
Source:
What have courts ruled with respect to AI and copyright? | Explained - The Hindu https://share.google/ESenkSwGp69Rgv3lz
Generated by ChatGPT AI app, July 25, 2025, upon request of Atty. Manuel Laserna Jr.