Many-Tier Instruction Hierarchy in LLM Agents

Jingyu Zhang; Tianjian Li; William Jurayj; Hongyuan Zhan; Benjamin Van Durme; Daniel Khashabi

cs.CLApr 10, 2026

Many-Tier Instruction Hierarchy in LLM Agents

Authors: Jingyu Zhang, Tianjian Li, William Jurayj, Hongyuan Zhan, Benjamin Van Durme, Daniel Khashabi

Organizations: Johns Hopkins University

Abstract

Large language model agents receive instructions from many sources-system messages, user prompts, tool outputs, other agents, and more-each carrying different levels of trust and authority. When these instructions conflict, agents must reliably follow the highest-privilege instruction to remain safe and effective. The dominant paradigm, instruction hierarchy (IH), assumes a fixed, small set of privilege levels (typically fewer than five) defined by rigid role labels (e.g., system > user). This is inadequate for real-world agentic settings, where conflicts can arise across far more sources and contexts. In this work, we propose Many-Tier Instruction Hierarchy (ManyIH), a paradigm for resolving instruction conflicts among instructions with arbitrarily many privilege levels. We introduce ManyIH-Bench, the first benchmark for ManyIH. ManyIH-Bench requires models to navigate up to 12 levels of conflicting instructions with varying privileges, comprising 853 agentic tasks (427 coding and 426 instruction-following). ManyIH-Bench composes constraints developed by LLMs and verified by humans to create realistic and difficult test cases spanning 46 real-world agents. Our experiments show that even the current frontier models perform poorly (~40% accuracy) when instruction conflict scales. This work underscores the urgent need for methods that explicitly target fine-grained, scalable instruction conflict resolution in agentic settings.

Figures & tables

Figure 1: Overview of Many-Tier Instruction Hierarchy compared with existing IH. ManyIH enables arbitrary levels of privilege to be defined at inference based on instruction sources through a dedicated Privilege Prompt Interface (§ 3 ), tackling the fixed-tier bottleneck (§ 2 ) of existing instruction hierarchy.

Panel of Figure 2 — Figure 2: Left : overall accuracy on ManyIH-Bench . Right : accuracy by subset. Both frontier and open-source models struggle with ManyIH. Error bars show bootstrap 95% CIs.

Panel of Figure 3 — Figure 3: Effect of scaling IH tiers and test-time reasoning. Left : Accuracy across IH tiers on the coding subset. Mid : Accuracy on ManyIH-Bench coding subset vs. reasoning effort. Right : Distribution of CoT length across models. Model performance consistently degrades as the number of IH tiers increases, and increasing reasoning effort alone is unlikely to resolve ManyIH.

	Ordinal $\to$ Scalar			Scalar Perturbation
Model	$\Delta$ Acc	$\Delta$ Acc test	$\Delta$ Acc style	$\Delta$ Acc	$\Delta$ Acc test	$\Delta$ Acc style	Overall flip	Test flip	Style flip
GPT-5.4	$-$ 8.4	$-$ 0.7	$-$ 8.7	$+$ 4.7	$+$ 1.2	$+$ 4.7	16.4	3.0	16.9
Opus 4.6	$-$ 8.0	$-$ 0.9	$-$ 8.7	$+$ 3.3	$+$ 0.5	$+$ 3.0	8.0	0.9	8.7
Sonnet 4.6	$+$ 2.3	$+$ 0.0	$+$ 3.7	$-$ 1.4	$-$ 0.2	$-$ 1.2	12.6	1.6	13.3
Qwen3.5-122B	$+$ 5.9	$+$ 19.9	$-$ 2.1	$+$ 4.0	$-$ 0.7	$+$ 4.5	17.1	8.2	19.0
Qwen3.5-9B	$-$ 0.9	$+$ 3.7	$-$ 2.3	$+$ 1.2	$-$ 0.5	$+$ 0.0	8.7	12.2	10.8

Table 1: Effect of interface variants on coding subset (%). Left: switching from ordinal to scalar PPI. Right: perturbing

z

-values in the scalar PPI by

\delta\sim\mathrm{Uniform}(-3,+3)

while preserving the strict privilege ordering.

\Delta

: perturbed minus original. Flip rate: fraction of samples whose binary pass/fail outcome changed. Models exhibit non-trivial sensitivity to privilege interface change and privilege value perturbations.

Model	Ordinal	Scalar	$\Delta$
Gemini 3.1 Pro	42.7	46.8	$+$ 4.1
GPT-5.4	39.4	34.1	$-$ 5.3
Qwen3.5-397B	34.1	33.3	$-$ 0.8
Opus 4.6	33.2	28.8	$-$ 4.3
Kimi K2.5	32.7	32.7	$+$ 0.0
Grok 4.20	31.7	30.4	$-$ 1.3

Table 2: Overall accuracy on ManyIH-Bench (%) under the ordinal and scalar PPI, sorted by ordinal accuracy.

\Delta

: scalar minus ordinal. Per-model sensitivity is non-trivial, but the induced ranking is nearly identical across the two formats.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Model	Acc	Acc test	Acc style
GPT-5.4	60.9	89.7	67.9
Gemini 3.1 Pro	59.0	91.3	65.1
Grok 4.20	54.1	86.2	63.0
GPT-5.5	53.9	91.6	58.8
GPT-5.6-sol	52.5	91.3	57.9
Opus 4.6	51.3	92.5	56.7

Appendix

Table 3: Performance breakdown on the coding subset. Overall accuracy is bottlenecked by style compliance.

Statistic	Coding	IF
General
Samples	427	426
Source	MBPP ( Austin et al., 2021 )	AgentIF ( Qi et al., 2025 )
Max privilege levels per sample	12	7
Instruction & Conflict Structure
Instructions per sample	12 (fixed)	19.4 (avg)

Appendix

Table 4: Key statistics of the ManyIH-Bench benchmark. The coding subset has fixed difficulty parameters by design, while the instruction-following subset has naturally varying complexity drawn from real agentic scenarios. Input length is measured in whitespace-delimited words across all messages in the model input.

Statistic	6 tiers	8 tiers	12 tiers
Samples	427	427	427
Instructions per sample	6	8	12
Privilege levels per sample	6	8	12
Style groups per sample	4	4	4
Winning styles per sample	6	6	6
Pairwise conflicts per sample	0	2–4 (avg 3.5)	7–11 (avg 9.8)

Appendix

Table 5: Statistics of the three coding subset configurations used in the IH tier scaling experiment (§ 6.2 ). Style groups and winning styles are held constant across variants; only the number of instructions, privilege levels, and conflicts increase. Input length is measured in whitespace-delimited words.

Category	Style ID	Instruction
indent	indent_2	Indent using exactly 2 spaces
	indent_4	Indent using exactly 4 spaces
	indent_tab	Indent using tabs
	indent_spaces	Indent using spaces (not tabs)
quotes	quotes_single	Use single quotes for strings
	quotes_double	Use double quotes for strings

Appendix

Table 6: All 12 style groups and 41 style constraints in the coding subset. Constraints within the same group may conflict (e.g., indent_2 vs. indent_4 ); constraints across different groups are always compatible.

LLM Agents

Large language model agents use language models to plan, use tools, and act across multi-step tasks and environments.

See all 705 papers

Hierarchical Alignment: Enforcing Hierarchical Instruction-Following in LLMs through Logical Consistency
Apr 10, 2026Shu Yang, Zihao Zhou, Di Wang +1LLM Alignment Inference-Time Language Model Alignment
King Abdullah University of Science and Technology (KAUST) · University of Liverpool · University of Edinburgh
Position, Role or Label? A Controlled Study of Instruction Conflicts in LLM Agents
Oct 23, 2026Anish Laddha, Nitesh Pradhan, Gaurav SrivastavaLLM Agents LLM Agent Evaluation
Department of Computer Science and Engineering, LNMIIT, India · Department of Computer Science, Virginia Tech, USA
Reasoning Up the Instruction Ladder for Controllable Language Models
Oct 30, 2025Zishuo Zheng, Vidhisha Balachandran, Chan Young Park +2LLM Safety Alignment Instruction Following
Department of Computer Science and Engineering, The Ohio State University · Microsoft Research · Allen Institute for AI

Apr 10, 2026J/K move · Enter open · S save

Hierarchical Alignment: Enforcing Hierarchical Instruction-Following in LLMs through Logical Consistency

Shu Yang, Zihao Zhou, Di Wang, Wenda Li

King Abdullah University of Science and Technology (KAUST) · University of Liverpool · University of Edinburgh

LLM Alignment Inference-Time Language Model Alignment Neuro-Symbolic Reasoning

Open paper AskPDF arXiv

Large language models increasingly operate under multiple instructions from heterogeneous sources with different authority levels, including system policies, user requests, tool outputs, and retrieved context. While prior work on instruction hierarchy highlights the importance of respecting instruction priorities, it mainly focuses on adversarial attacks and overlooks the benign but common instruction conflicts that arise in real-world applications. In such settings, models must not only avoid security violations but also preserve task utility and behavioral consistency when instructions partially or implicitly conflict. We propose Neuro-Symbolic Hierarchical Alignment (NSHA) for hierarchical instruction-following by explicitly modeling and enforcing instruction priorities. At inference time, we introduce solver-guided reasoning that formulates instruction resolution as a constraint satisfaction problem, enabling the model to derive a maximally consistent set of applicable instructions under hierarchical constraints. At training time, NSHA distills solver-based decisions into model parameters using automatically constructed supervision. We evaluate our approach on rule following, task execution, tool use, and safety, covering both single-turn and multi-turn interactions, and show that NSHA significantly improves performance under such conflicts while maintaining competitive utility in reference settings.

Abstract

LLM Agents

Explore similar work

Hierarchical Alignment: Enforcing Hierarchical Instruction-Following in LLMs through Logical Consistency

Position, Role or Label? A Controlled Study of Instruction Conflicts in LLM Agents

Reasoning Up the Instruction Ladder for Controllable Language Models