Abstract Generative artificial intelligence has unsettled an assumption that the law of intellectual property had never before been forced to examine: that behind every protected work there stands a human creator. This paper argues that the disturbance operates at two distinct points, and that the failure to keep them apart is the source of much confusion in the present debate. At the output, the machine produces work for which no human author can readily be identified, and the law must decide whether, and on what terms, such work may be protected. At the input, the same machine is built by copying, on an unprecedented scale, the protected work of the very human creators the law professes to serve. Examining the recent authorities the refusal of registration in Thaler v. Perlmutter and the United States Supreme Court’s denial of certiorari, the Indian Copyright Office’s irresolute handling of the RAGHAV application, the DABUS patent litigation, and the wave of training-data suits led by Andersen and The New York Times the paper contends that the human-authorship requirement is structural rather than incidental, that wholly autonomous output therefore falls outside the existing law by design and not by oversight, and that India’s closed list of fair-dealing exceptions leaves the training of models markedly more exposed than the open-ended fair-use doctrine of the United States. It proposes, for India, a calibrated response: the denial of protection to autonomous output, the protection of genuinely AI-assisted human creation by a test of meaningful creative control drawn from existing doctrine, and a transparent, remunerated licensing regime for the use of protected work in training. Keywords: artificial intelligence; copyright; authorship; originality; training data; fair dealing; Section 2(d)(vi); intellectual property reform.