Intelligence systems fail not when they produce errors but when errors enter feedback loops that compound rather than correct. This paper introduces the Recursive Hallucination Principle: any sufficiently autonomous intelligence system that recursively consumes its own outputs without adequate external verification will tend to diverge from reality over time. The principle is not specific to artificial intelligence. Financial bubbles, organisational groupthink, and media ecosystems all demonstrate the same structure. Artificial intelligence did not create recursive hallucination. It industrialised it — enabling recursive feedback loops to operate at machine speed, machine scale, and with machine persuasiveness. Without verification, recursion becomes amplification. With verification, recursion becomes learning. The distinction between the two is the distinction between intelligence that compounds and intelligence that collapses. ---