India’s Copyright Office has drawn the same line US regulators and courts have now settled on: an AI system cannot be listed as the author of a work, no matter how sophisticated the output. According to The Indian Express, the ruling tracks the reasoning behind Thaler v. Perlmutter, the case the US Supreme Court declined to review on March 2, 2026, leaving intact a D.C. Circuit opinion that affirmed the US Copyright Office’s human-authorship requirement, according to Mayer Brown. The practical map is straightforward — companies that treat AI as a tool inside a documented human creative process keep something to license; companies betting on fully autonomous generation as an ownable asset class do not.
The US Copyright Office’s January 29, 2025 Part Two report, which drew more than 10,000 public comments according to Skadden, is doing the heavy lifting behind this convergence. It sorts AI use into four buckets — facilitating a human creative process, prompting alone, expressive inputs, and human arrangement of AI outputs — and only the first, third, and fourth categories clear the bar. Prompts, the Office concluded, “function as instructions that convey unprotectable ideas,” per Mayer Brown’s summary, which means the more a workflow looks like typing a request and taking the first result, the less defensible it is as intellectual property.
Raw AI output is functionally entering the market unprotected — which is good news for anyone downstream who wants to reuse it, and bad news for anyone upstream who wanted to sell exclusivity on it.
That asymmetry is the real market signal here. Sellers of stock imagery, editorial content, and licensed datasets who can show a documented human hand — selection, arrangement, editing, iterative refinement — retain a moat that pure-play AI content generators cannot replicate. Buyers of licensed data and content, in turn, now have a real incentive to demand provenance documentation as a contract term, not a courtesy; expect compliance and audit-trail tooling around “human contribution” records to become its own small vendor category. Meanwhile, the fact that unadorned AI output carries no copyright at all cuts the other way for data aggregators and AI trainers, who face less infringement exposure scraping AI-generated material precisely because there’s no exclusive right attached to it — even as courts, per the Thomson Reuters v. Ross Intelligence ruling flagged by the New York State Bar Association, keep finding real teeth in copyright claims over human-authored training data used without a license.
Watch Allen v. Perlmutter, the pending Colorado case testing whether 600-plus iterative prompts can cross into authorship, and the US Copyright Office’s still-unreleased Part Three report on training-data licensing and liability. Both will tell data companies exactly where the documentation bar sits — and, now that India appears to be reading from the same rulebook, whether that bar becomes a de facto global standard rather than a US quirk.
AI can create a work, but can it be its author? Why the Copyright Office drew the line
