COGNITIVE EXTRACTIVISM AND ALGORITHMIC SOVEREIGNTY

the expropriation of citizen's intellectual property by global LLMs

Abstract

This article investigates the phenomenon of cognitive extractivism driven by the mass adoption of Large Language Models (LLMs) by ordinary citizens. The objective is to analyze how the usage policies of major proprietary Artificial Intelligence (AI) systems address the intellectual property entered in their prompts and to assess the resulting risks to algorithmic sovereignty. The method employed consists of an automated audit, using the Retrieval-Augmented Generation (RAG) technique, of the Terms of Service and Privacy Policies of Google (Gemini), OpenAI (ChatGPT), and DeepSeek. To this end, the ESA-LLM framework (Extraction of Sovereignty and Authorship) was developed, composed of variables related to licensing, data memorization, and jurisdiction. The results reveal a profound contractual asymmetry: while the authorship of the output is nominally assigned to the user, the input (ideas and raw knowledge) is expropriated for the continuous training of the platforms. It is concluded that the user relinquishes their data sovereignty in favor of transnational infrastructures governed by the laws of California (USA) or Hangzhou (China), demonstrating that current data protection legislation is insufficient to safeguard national intellectual capital against algorithmic memorization.