Researchers developed U-Fuzz, a fuzzing tool to systematically discover memory-use failures in LLM agents. U-Fuzz targets query-related and memory-state failures in persistent memory, improving the detection of memory-use errors in LLMs across various memory architectures.