This conceptual essay introduces the notions of pre-expressive freedom, pre-expressive censorship, post-visible cage, and cognitive sovereignty as analytical tools for understanding the governance of frontier artificial intelligence. The central hypothesis is that when AI systems become cognitive infrastructures, the control of access to models, filters, classifiers, memory, and authorization regimes is no longer merely a technical or commercial matter. It becomes a political question concerning the conditions under which human thought can be formulated, investigated, expanded, and expressed. Traditional freedom of expression focuses on the right to speak, publish, criticize, and circulate ideas. However, in an AI-mediated environment, expression is increasingly preceded by algorithmic processes of drafting, inquiry, simulation, translation, reasoning, and co-formalization. If these pre-expressive processes are governed by opaque classifiers, differentiated access regimes, safety policies, compliance systems, or corporate and state interests, then freedom may be constrained before public expression occurs. The essay does not deny the legitimacy of AI safety or regulation. On the contrary, it recognizes that high-capability AI systems generate real risks. The key concern is whether safety can become a totalizing language through which access to cognitive power is asymmetrically administered. The paper argues that cognitive sovereignty should become a central concept in AI governance, requiring transparency, contestability, plural model access, independent auditing, and safeguards against both corporate and state capture.
Joaquim Santos Albino (Wed,) studied this question.