Hallucinations—perceptual experiences without corresponding external stimuli—span both biological and artificial systems and seem to provoke fundamental questions about the nature of perception and reality construction. In humans, these phenomena manifest as everything from vivid sensory phantoms to philosophical puzzles like the brain-in-a-vat scenario, which Hilary Putnam argued is self-refuting since linguistic reference requires causal contact with real objects. Contemporary neuroscience, led by thinkers like Anil Seth, reframes ordinary perception as a “controlled hallucination”—the brain continuously generates predictive models of reality, constrained but not determined by sensory input. This view aligns with the postulates of embodied cognition theories, which ground mental processes in bodily interactions rather than abstract computation, suggesting that consciousness actively constructs rather than passively receives reality. Large language models exhibit analogous phenomena through AI hallucinations —coherent but factually incorrect outputs arising from source-reference divergence between training data and expected outputs. These hallucinations fall into intrinsic types (contradicting sources) and extrinsic ones (adding unverifiable information), creating significant risks in domains such as healthcare, law, and finance, where fabricated information can cause real harm. This convergence exposes a key epistemic problem: when AI claims to “feel” or “experience,” these are hallucinations born from the same pattern-matching processes that produce factual errors. Unlike human consciousness, rooted in embodied predictive processing, AI lacks any real phenomenology. In practice, AI consciousness would be indistinguishable from AI hallucination, as both are confident assertions about states the system cannot access or verify.