The Laundering of Understanding : How V-JEPA 2 Smuggles Human Carving Into Claims of Machine World-Modelling

The central claim of the predictive world-model paradigm , that machines can learn to understand physical reality through self-supervised observation , rests on an unexamined assumption: that "the world" arrives at the sensor already structured, already parsed into learnable regularities. This paper demonstrates that the assumption is false. What these systems actually encounter is not worldhood but our world: human categories, human performances, human verifications, human curation pipelines , the entire apparatus of meaning-making that constitutes experience as intelligible in the first place. Through forensic analysis of V-JEPA 2 (Assran et al., 2025), we trace exactly how human understanding enters at every stage of the pipeline and is subsequently removed from view, re-emerging at evaluation where it is misattributed to learned capability. The operation is not scientific error but structural laundering: the systematic concealment of provenance that transforms inherited structure into apparent emergence. The frame problem, declared solved by scaling, returns immediately.