Huge language models erase brand identity while keeping usefulness
LLM unbranding: Erasing Commercial Identity while Preserving Generic Utility
Computation and Language
Summary
Large language models often mention brands in ways that can cause problems like confusing trademark ownership or damaging brand reputations. The paper defines a new challenge called LLM unbranding, which means removing brand-specific words and phrases from texts without hurting the model's general abilities. The authors created a benchmark dataset to test current methods and found they struggle to remove brand identity properly. They then developed a new method called MUTE that safely removes brand references during inference without changing the model itself.
What this means in practice
- •For ai developers: Improve language model outputs by removing brand-specific language while keeping responses generally useful and accurate.
- •For content moderation teams: Reduce legal and reputational risks by filtering out brand trademarks and slogans from AI-generated text.
- •For advertising agencies: Automatically neutralize brand names and style in generated content to avoid unintentional brand dilution or defamation.$Commercial implications: Provides a tool for agencies to safely manage brand mentions in AI-created marketing materials.
Authors
Kajetan Ożóg, Alicja Wojciechowska, Dawid Malarz, Paweł Batorski, Artur Kasymov, Przemysław Spurek
Abstract
Establishing unbranding as a critical practice to prevent visual logos from acquiring negative connotations is standard in image generation. Large Language Models (LLMs) now face a parallel and emerging challenge. These models frequently generate brand descriptions within diverse contexts. This frequency introduces significant risks, such as trademark dilution, false attribution, and brand defamation. In response, we formally define the novel task of LLM Unbranding. We specifically address the complex challenge of managing trade dress within textual outputs. This involves neutralizing characteristic language, slogans, and stylistic markers that define brand identity. Crucially, these elements are less evident than explicit visual logos. To benchmark this task, we introduce a comprehensive evaluation dataset incorporating prominent brands from multiple commercial domains. We rigorously evaluate existing state-of-the-art machine unlearning models using this benchmark. This evaluation identifies their limitations in selective textual unbranding. Finally, we propose MUTE, a novel inference-time method that effectively neutralizes textual trade dress while preserving the LLM's general capabilities and utility. By leveraging an iterative refinement loop, MUTE systematically optimizes system instructions to safely eliminate brand leakage without requiring fragile parameter updates. Code and dataset: The evaluation dataset and code for LLM Unbranding are available at https://github.com/KajetanOzog/LLM_unbranding. The implementation of MUTE is available at https://github.com/KajetanOzog/MUTE.