Abstract: Machine translation hallucination refers to the phenomenon that the translated text contains information unrelated to the source language sentence. This paper proposes a fact-verification-based five-chain machine translation hallucination detection agent. This agent uses five sequential processes to determine whether the facts in the machine translation align with those in the source language, thus enabling the detection of hallucination phenomena in machine translation. Experimental results on the HalOmi sentence-level high-resource natural hallucination detection subtask show that the proposed method significantly improves the performance of hallucination detection and outperforms the best competing methods for most language pairs.