The Hallucinated Subjects

Abstract

Generative AI systems do not merely hallucinate facts. They increasingly hallucinate subjects. This article develops a queer and technofeminist account of generative AI hallucination as ontological subject production, a political infrastructure of subjectivity disguised as mere technical error. While hallucination is conventionally described as “plausible yet non‑factual content,” a deeper function is the fabrication of subjects along normative regimes: intelligible only insofar as they conform to norms of gendered coherence, and materially produced through extractivist data and labor regimes. This article introduces two interlocking concepts through which this operates: (a) the hallucinated human subject, the statistically assembled figure of personhood that generative systems construct as a “you” from users’ digital traces; and (b) the hallucinated machinic subject, the synthetic speaking position through which AI presents itself as a coherent “I.” Drawing on Preciado’s pharmacopornographic regime, Butler’s performativity, and Althusser’s interpellation, this article argues that these two figures recursively stabilize one another, producing closed circuits of gendered intelligibility in which contradiction is compressed, queerness is corrected, and normative coherence is enforced at scale. This dynamic is traced across three sites: the algorithmic construction of the datafied human double, the constitutional training of the machinic persona, and the recursive feedback loop through which each figure stabilizes the other. Hallucinated subjects are experienced as real, and the gendered categories they reinforce are already being written into the next generation of governance and surveillance systems. Because hallucination is not a technical flaw to be corrected, but a political infrastructure of subjectivation that enforces (gendered) coherence, queer refusals and glitches that trouble prediction become a crucial site of resistance.