Delhi HC sides with OpenAI: AI copyright law is now local
India's Delhi High Court denied ANI interim relief, calling LLM training fair dealing — days after a US court approved a $1.5B settlement. AI copyright now varies by market.
Two courts, one month, opposite answers. On July 24 the Delhi High Court refused ANI's request to restrain OpenAI, holding that training ChatGPT on the news agency's content is not, at this stage, copyright infringement. Weeks earlier a US judge signed off on Anthropic's $1.5 billion settlement over training data. Same technology, same legal question, wildly different price tag. AI copyright law is not converging — it's fragmenting by jurisdiction, and that's a planning input if you publish anything on the internet.
What actually happened
ANI Media sued OpenAI and asked for an interim injunction. The court said no on essentially every ground.
Training on copyrighted material, the court held, falls prima facie within the fair dealing exception in Section 52(1)(a) of India's Copyright Act, 1957. Storing ANI's works, per the order, does not by itself amount to infringement. And ANI failed to establish a prima facie case that ChatGPT reproduced its reporting — outputs produced through retrieval-augmented generation weren't found substantially similar to the originals. The court also weighed the balance of convenience toward OpenAI, reasoning that halting operations at the interim stage would cause irreparable harm to the defendant and to broader public interest in AI development, as Free Press Journal reported.
One caveat the court made itself: these findings are limited to the interim application and don't decide the merits. Nothing here is final.
Why it matters for your business
Two practical reads.
First, your published content is training data, and what you can do about it depends on where you'd sue. A US class action extracted a nine-figure settlement. An Indian court just declined to even pause the same category of activity. If your protection strategy was "eventually the law will sort this out," it won't — not uniformly. The durable moves are the ones that don't depend on a courtroom: content that carries your operating experience rather than restatable facts, distribution you own, and a customer relationship a model can't summarize away.
Second, this is vendor risk in both directions. Copyright exposure is one of the real overhangs on frontier model pricing, and it now resolves differently in every market your vendor operates in. A model that gets enjoined or repriced in one jurisdiction while running fine in another is exactly the scenario that makes a swappable model layer worth building. Keep prompts, evals, and routing in your own code so switching providers is a config change — not a rebuild.
Key takeaways
- Delhi HC denied ANI interim relief July 24, calling LLM training prima facie fair dealing under Section 52(1)(a)
- Court found storage isn't infringement and RAG outputs weren't substantially similar to ANI's reporting
- Findings are interim only — the case is not decided
- AI copyright liability now varies sharply by jurisdiction; don't build a content or vendor strategy that assumes one global answer
Building on a model whose legal footing keeps moving? We build AI systems with the model layer swappable, so a court order or price change in one market is a config change, not a rewrite. See what we've shipped or talk through your stack.
Sources: LiveLaw, Free Press Journal.
- #ai-copyright
- #openai
- #regulation
- #training-data
- #vendor-risk
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Pilot Protocol's $4.5M: a directory isn't a standard
A seed-stage AI agent network wants to be where agents discover each other. Useful, but know the difference between an open spec and someone else's front door.
Read itNvidia backs Safe Superintelligence, discloses no terms
Nvidia's Safe Superintelligence partnership names no dollar figure and no term length. Reported at $5B. Here's how to read AI deals that omit the numbers.
Read it