{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/control-illusion-the-failure-of-instruction","title":"Control Illusion: The Failure of Instruction Hierarchies in Large Language Models","arxiv_id":"2502.15851","date":"2025-02-21","proceeding":null,"authors":["Yilin Geng","Haonan Li","Honglin Mu","Xudong Han","Timothy Baldwin","Omri Abend","Eduard Hovy","Lea Frermann"],"abstract":"Large language models (LLMs) are increasingly deployed with hierarchical instruction schemes, where certain instructions (e.g., system-level directives) are expected to take precedence over others (e.g., user messages). Yet, we lack a systematic understanding of how effectively these hierarchical control mechanisms work. We introduce a systematic evaluation framework based on constraint prioritization to assess how well LLMs enforce instruction hierarchies. Our experiments across six state-of-the-art LLMs reveal that models struggle with consistent instruction prioritization, even for simple formatting conflicts. We find that the widely-adopted system/user prompt separation fails to establish a reliable instruction hierarchy, and models exhibit strong inherent biases toward certain constraint types regardless of their priority designation. While controlled prompt engineering and model fine-tuning show modest improvements, our results indicate that instruction hierarchy enforcement is not robustly realized, calling for deeper architectural innovations beyond surface-level modifications.","url_abs":"https://arxiv.org/abs/2502.15851v1","url_pdf":"https://arxiv.org/pdf/2502.15851v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"control-illusion-the-failure-of-instruction","repo_url":"https://github.com/yilin-geng/llm_instruction_conflicts","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"prompt-engineering","task_name":"Prompt Engineering"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/2502.15851","atlas_url":"https://app.syntology.ai/?focus=2502.15851","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}