Taint analysis plays a crucial role in fuzzing by identifying input bytes that significantly influence program behavior. However, existing taint analysis approaches either require heavy- weight instrumentation or incur substantial runtime overhead. In this paper, we propose HFuzz, a novel fuzzing approach that leverages the havoc mutation mode - a fundamental component in modern fuzzers - to perform lightweight taint inference. Our approach operates in two phases: first identifying ”hot bytes” through havoc-based sampling, then using this information to guide subsequent mutations. By utilizing existing fuzzing components rather than adding extra execution, HFuzz achieves efficient taint inference while maintaining the simplicity and scalability of conventional fuzzers.