The Impossibility of AGI Safety

This volume draws on Rice's theorem, dynamical systems theory, and game-theoretic analysis to construct multiple independent impossibility arguments. It addresses and rebuts proposed alignment mechanisms including error correction regress, architectural privilege, delegation, and cooperative approaches. The volume also examines physical impossibility arguments related to substrate incompatibility and presents quantitative bounds on the entropy growth that prevents sustained alignment over time. The only certain path to safety is not to build AGI because the upside is uncertain, bounded, and temporary while the downside is catastrophic, irreversible, and possibly total, making development an asymmetric bet with negative expected value for any nonzero probability that the arguments are correct. The recommended actions are to not design, build, or deploy AGI, to not deceive others about AGI safety, to support institutional recognition of the impossibility, and to treat AGI development as maximally irresponsible, analogous to building doomsday devices.