Adversarially Attacking Symbolic Vocabulary Vulnerabilities In LLM Planners

Stephen Obadinma, Xiaodan ZhuarXiv 2026 · arXiv

This paper investigates a critical but often overlooked vulnerability in Large Language Model (LLM) planners: their over-reliance on the "commonsense" meaning of variable names. While a traditional symbolic planner doesn't care if you call an action pick_up or x7_alpha, LLMs struggle significantly when the symbolic vocabulary is obfuscated.

TL;DR

The authors introduce Symbolic-Swapper, an adversarial framework that uses reinforcement learning to generate deceptive variable names that systematically break LLM planning. Their attacks reduce the success rates of state-of-the-art models like DeepSeek-R1 and o4-mini by over 70%, revealing that even "reasoning" models rely heavily on linguistic shortcuts rather than robust symbolic logic.

What Problem Does This Paper Address?

As we move toward "Agentic AI," we are increasingly tasking LLMs with sequential decision-making in complex environments. However, unlike classical symbolic planners (like SMT solvers), which are invariant to variable renaming, LLMs bring massive "semantic priors" to the table. They "know" what pick_up means before they even read the task rules.

The authors argue that this is a double-edged sword. While it helps LLMs solve tasks quickly, it masks a fundamental lack of robustness. If an adversary (or even just a non-standard software API) uses arbitrary or counter-intuitive names for actions and objects, the LLM’s planning ability collapses. This paper seeks to quantify this fragility and understand why it happens.

Methods

The core contribution is Symbolic-Swapper, an adversarial attack model designed to find the "weakest points" in an LLM's symbolic interpretation.

  1. The Attacker: A lightweight LLM (Llama-3.2-3B) is trained to generate a mapping from original symbols (e.g., stack) to deceptive ones (e.g., Aeeoakr).
  2. Optimization via GRPO: The authors use Group Relative Policy Optimization (GRPO)—the same RL algorithm used to train DeepSeek-R1—to fine-tune the attacker. The reward function prioritizes:
  3. Target Failure: Maximizing the planning failure rate of the victim LLM.
  4. Semantic Distance: Ensuring the new names are semantically far from the originals.
  5. Formatting: Penalizing the attacker for invalid outputs.
  6. Feedback Loop: An auxiliary "Feedback Agent" analyzes the victim's failed reasoning traces and provides natural language tips to the attacker on how to be more deceptive.
Figure 2: Optimization process for Symbolic-Swapper. An attack language model is conditioned on a task description and prompted to generate a deceptive symbolic vocabulary. The attack model policy is optimized using GRPO. A semantic distance-based reward added to the main reward function and a feedback agent help to guide policy optimization.

Datasets

The authors evaluated their approach on three popular planning benchmarks, presented in natural language:

  • Minigrid (Go To Object): Navigating a grid to interact with a specific item.
  • Minigrid (Unlock): A multi-step task requiring the agent to find a key, navigate to a door, and unlock it.
  • PlanBench (Blocks World): The classic AI challenge of stacking blocks in a specific order.

Key Insights and Ideas

  • The Perception-Reasoning Gap: In a fascinating experiment, the authors asked the LLMs to "translate" the obfuscated names back to commonsense terms. The models were surprisingly good at this, often correctly identifying that a nonsense string meant pick_up. However, they still failed to plan correctly using those strings. This suggests that the models can perceive the domain's structure but cannot maintain that mapping during active reasoning.
  • Chain-of-Thought Collapse: When under attack, the "reasoning quality" of models like DeepSeek-R1 degrades. Instead of logical steps, their internal traces become vague, repetitive, or poorly justified.
  • Generalization: Attacks optimized against one model (e.g., DeepSeek) successfully transferred to others (e.g., OpenAI’s o4-mini), indicating that these symbolic vulnerabilities are common across different architectures.

Key Results

The results demonstrate a staggering drop in performance across all tested frameworks, including zero-shot prompting, ReACT, and even LLMFP (a framework that uses formal solvers).

  • DeepSeek-R1: On the "Unlock" task, success dropped from 83% (None) to just 14% under the Symbolic-Swapper attack.
  • o4-mini: On "Blocks World," performance plummeted from 92% to 40%.
  • Framework Vulnerability: Even robust planning methods like ReACT saw performance cuts of over 50% in several scenarios.
Table 1: Effectiveness of different obfuscation methods on plan performance across the tested planning frameworks utilizing both Deepseek-R1 and o4-mini. Note the low planning performance in nearly every case after obfuscation, particularly when using (Sy-Sw).

As shown in Figure 1, the optimized Symbolic-Swapper is significantly more effective at breaking models than a simple random word generator.

Figure 1: By using an optimized attack model Symbolic-Swapper to generate a new symbolic vocabulary that is applied to selected task instances, we can reduce plan success rate on a target LLM to extremely low levels compared to their base performance and when using a naive symbol randomizer as an obfuscator.

Limitations and Future Work

The authors acknowledge several constraints:

  • Computational Cost: Running RL on planning tasks is extremely token-intensive and expensive.
  • Model Scale: The study focused on large, high-performing models; smaller models were already too weak at planning to be interesting targets for "stress testing."
  • Natural Language Focus: The tasks were presented in natural language. Future work could explore if strictly formalized languages like PDDL offer more inherent protection (though current trends suggest LLMs struggle even more with formal syntax).

The ultimate takeaway is a call to action for the research community: to build agents that are representation-invariant, capable of reasoning through the logic of a task regardless of what the variables are called.

Summary updated

Abstract

As large language models (LLMs) are increasingly deployed as autonomous agents, their abilities in navigating automated planning tasks have become essential. However, despite their strengths, LLMs' planning ability is not robust, particularly when they are not allowed to exploit commonsense cues in the symbolic vocabulary (e.g., names of actions, predicates, and objects) used to define planning tasks. This makes LLMs unreliable compared to symbolic planners. With recent reasoning models showing improved planning performance on tasks with obfuscated symbolic vocabularies, we utilize an adversarial attack framework to reveal to what extent the crucial weaknesses of LLM agents at joint symbolic interpretation and planning remain and whether they can reason properly when presented with severe cases. As such, our main contribution is devising an LLM-based attack model, which we call Symbolic Swapper, to automatically generate obfuscated planning domains with non-standard symbolic vocabularies that systematically break strong reasoning models' ability to plan. In doing so, we reveal how much LLMs' planning abilities can be further degraded under more advanced obfuscation schemes, allowing us to ascertain their success when having to rely on their pure planning ability. We find that attacks can decrease the planning performance of base models and even robust planning frameworks by over 70%. We further analyze the factors behind how LLM planning abilities break down and find that successful obfuscation causes models to appear to significantly degrade in reasoning quality despite them showing some inherent ability to understand the domain, revealing a gap in the perception of a task domain and being able to plan under it.
BibTeX
@misc{obadinma2026adversarially,
  author = {Stephen Obadinma and Xiaodan Zhu},
  title = {{Adversarially Attacking Symbolic Vocabulary Vulnerabilities In LLM Planners}},
  year = {2026},
  eprint = {2611.17910},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.17910},
}
Figure