Benchmarking Safety Risks of Knowledge-Intensive Reasoning under Malicious Knowledge Editing | ArxivCSExplorer