Researchers propose an energy landscape analysis framework to explain why jailbreak attacks on diffusion language models (dLLMs) succeed. They identify two strategies used by attacks and derive three detection signals that can identify malicious queries.