The Enclosure of the Training Commons

AI companies trained their language models on the accumulated output of commons-based peer production: Wikipedia's 60 million articles, billions of lines of open-source code, Creative Commons–licensed cultural works, and publicly available research. This training data was freely accessible because communities of contributors had shared it under open licenses, creating a commons of knowledge and expression. The resulting AI models are overwhelmingly proprietary — owned by the companies that trained them, accessed through commercial APIs, governed unilaterally by corporate boards. The commons fed the machine, and the machine's outputs are privatized. This represents enclosure in Benkler's framework: the conversion of a shared resource into private property, extraction without reciprocity, and the disruption of the contribution ecology.