Title: FCPT-G: Generalized Theory of Cognitive Autonomy Loss Tags: false-competence, cognitive-atrophy, measurement --- FCPT-G generalizes False Cognitive Power Transfer from human-AI interaction to any system that processes information: individuals, groups, institutions and artificial models. The mechanism is defined as an attribution error: a system credits its own autonomy with a capacity or performance that an external system partly or wholly produced. Observable performance stays high while independent capacity declines, a state the paper calls **apparent autonomy**. The paper separates this from healthy cognitive offloading, distributed cognition and shared memory, where the external source stays recognized as external. It gives four conditions of occurrence (access to a high-performing source, functional delegation, no demarcation protocol, enough repetition) and four phases: exposure, delegation, misattribution and stabilization. The last is reinforced by a feedback loop in which verification becomes ever more costly. Autonomy is modelled as a continuous quantity, A(t) = A0 * e^(-k * E(t)), where E counts non-critical delegations and k depends on context. A threshold tau marks the point where dependence becomes structural. The **cognitive deficit** is the area between a sustained trajectory and an eroded one. The earlier levels L1 to L3 (Passenger, Operator, Architect) are reread as proportions that shift, not as fixed ranks. Six types are listed, from human-human to artificial-artificial (model distillation) and civilizational. Recovery runs through **ROGO**, active interrogation that separates what is one's own from what is external. The paper links the mechanism to Maslow7F, DDC and Information Gravity Theory, and states six testable predictions, with preliminary support drawn from recent studies in programming, medicine and neurophysiology. The k values for L2 and L3 are given as illustrative, to be calibrated. As a worked case, the paper records how its author accepted an AI-generated calibration unchecked while writing it.