Abstract. Despite the wide availability of automated testing techniques such as fuzzing, little attention has been devoted to testing computer architecture simulators. We propose a fully automated approach for this task. Our approach uses large language models to create input programs, including information about their parameters and their types, as test cases for the simulators. The LLM’s output becomes the initial seed for an existing fuzzer, AFL, which has been enhanced with three mutation operators, targeting both the input binary program and its parameters. We implement our approach in a tool called SearchSYS. We use it to test the gem5 system simulator. SearchSYS discovered 21 new bugs in gem5, 14 where gem5’s software prediction differs from the real behaviour on actual hardware and 7 where it crashed. New defects were uncovered with each of the 6 LLMs used. SearchSYS tool is available at https://github.com/karineek/SearchGEM5/. Experiments in this publication were done with commit 6514a24. The previous publication of SearchSYS (SearchGEM5), is available here: https://zenodo.org/records/10999115. This record includes all data collected during the experiments between January to August 2024. The code: ASEGem5-main.zip. gem5 instrumented with coverage and gcc-9: gem5-instrumented-with-coverage-x86-ubuntu-20.04-gcc-9.zip. The Docker in the coverage folder in ASEGem5-main.zip works for more recent gcc versions. The data of Experiment 1: Initial corpus files (RQ1): LLM_test_inputs.zip. Bugs analysis: BugHuntingData_exp1_raw.tar.gz Coverage of initial corpus (RQ1-2): LLMresultsCov_beforeFuzzing.zip. Configurations results (RQ3): Experiment-1-1h-30_repeat_30_confg.zip (data) and Experiment-1-AnalysisRes-Select-Values4Counters.xlsx (analysis) The data of Experiment 2 (RQ4): Data after fuzzing (fuzzed corpus): -exp2.tar.gz Coverage after fuzzing: CodeCoverage-Main-PostAFL.zip Analysis of the data: exp2-data.zip (raw) and gem5_mism_bugs_29-August_2024.xlsx (excel)