Can a copyright law written in 1957, long before anyone imagined a machine reading millions of articles overnight, meaningfully decide what an AI company is allowed to do with a newsroom's work in 2026?

The Real Legal Move Here Isn't "AI Won"

• It's a 1957 Word Being Redefined Here's what's actually happening. Everyone is reading this as "court sides with OpenAI." That's the headline, not the mechanism. The actual legal move is something called the doctrine of updating construction - the idea that an old law's words can be read to cover situations Parliament never imagined, as long as the underlying purpose still fits. Section 52(1)(a) of the Copyright Act, 1957 says using someone's work for "research" isn't infringement. In 1957, "research" meant a person in a library, taking notes. Justice Bansal has now read that same word to cover a machine statistically absorbing patterns from millions of articles to build a language model. That's not a small stretch - it's the court deciding that "research" is a purpose, not a method and that a 69-year-old sentence can hold a technology built decades after the sentence was written.

Why the Court Didn't Stop at "It's Research"

Two more pieces had to fall into place and both matter more than they look. First: ChatGPT's outputs weren't found to be "substantially similar" to ANI's actual articles - meaning the model learned patterns, not verbatim text.

If it had regurgitated ANI's copy word-for-word, the fair dealing defence collapses instantly. Second: the court noted ANI's own 7.5-million-dollar licensing offer to OpenAI, which the judge used to argue the harm is measurable in money, not the kind of irreversible damage that justifies blocking a product mid-development.

Where India Fits Globally

This isn't India inventing a position from scratch - it's catching up to a pattern. In June 2025, two US federal courts, ruling on cases against Anthropic and Meta within days of each other, reached the same underlying conclusion: training an AI model is "transformative," not a substitute for the original work, so it doesn't dent the author's market.

Delhi HC leaned directly on both. India is, for now, aligning with the US on the permissive end of this global spectrum - not writing new law, but importing an interpretive move that's already spreading.

What This Doesn't Settle This is an

interim order, not a final judgment - the underlying case is still alive. And the "not substantially similar" test only protects AI companies as long as outputs stay paraphrased. If a future model starts reproducing near-verbatim text, this exact reasoning stops applying. The ruling protects the training step, not every possible output. For the exam, the sharp point isn't "AI training is legal in India"

• it's that courts are now the ones deciding, sentence by sentence, how far old statutory language stretches to cover technology its drafters never saw coming.