TextCloak: Thwarting Unauthorized LLM Exploitation via RL-Driven Unlearnable Text

2026-07-30

AI Summary

This work proposes TextCloak, the first framework for generating unlearnable text in open-ended generation scenarios to prevent unauthorized fine-tuning of large language models (LLMs) on user-provided content. Departing from conventional approaches that rely on class labels, TextCloak employs a bilevel optimization scheme coupled with a generative policy network to produce protective texts that effectively disrupt illicit fine-tuning while preserving semantic fidelity and linguistic naturalness. Guided by performance degradation signals from downstream proxy models via the GRPO-UE algorithm, TextCloak adaptively optimizes its perturbations. Extensive experiments across six datasets and nine prominent LLMs demonstrate that TextCloak substantially degrades the efficacy of unauthorized fine-tuning without compromising text utility, while also exhibiting strong cross-model generalization and robustness against adaptive attacks.

Abstract

The rapid development of Large Language Models (LLMs) has led to significant advances across a wide range of language tasks, while simultaneously raising growing concerns about unauthorized data exploitation and privacy leakage. Unlearnable examples (UEs) offer a promising defense by introducing carefully designed perturbations into data such that models trained on them exhibit degraded utility. However, existing methods for text protection are primarily designed for classification tasks (e.g., sentiment analysis) in discriminative language models and often rely on injecting class-specific linguistic cues, which limits their effectiveness in the open-ended generation settings of LLMs. In this work, we propose TextCloak, an RL-driven framework for protecting textual data against unauthorized LLM exploitation. TextCloak employs a generative policy that transforms batches of clean text into unlearnable examples while preserving semantic fidelity and linguistic naturalness. To optimize the policy, we introduce GRPO-UE, which rewards generated unlearnable text based on the downstream degradation they induce in fine-tuned surrogate LLMs and updates the generator parameters via group-relative policy optimization. This bi-level optimization enables the generator to discover generalizable protective patterns beyond class-specific cues. Comprehensive experiments on six publicly available datasets and nine state-of-the-art LLMs demonstrate that TextCloak consistently impairs unauthorized fine-tuning while maintaining text utility for legitimate use. Further analyses establish its transferability and robustness across model architectures, training configurations, and adaptive attacks, highlighting its broad applicability as a practical defense against unauthorized LLM exploitation.

Problem

• unauthorized LLM exploitation
• unlearnable examples
• text protection
• privacy leakage
• large language models

Innovation

• Unlearnable Examples
• Reinforcement Learning
• Large Language Models
• Text Privacy
• Adversarial Defense