On the Vulnerability of Parameter-Level Defenses to Model Merging
The training-free integration of expert models via model merging has exposed significant security risks, enabling free-riders to combine specialized models without authorization. Recent works propose parameter-level defenses that employ linear parameter transformations to neutralize this threat. In this paper, we systematically analyze such defenses and reveal that their protected task vectors are inherently small in magnitude. Consequently, the protected weights remain overwhelmingly dominated by the pretrained model. Based on this observation, we designate the pretrained model as a static reference anchor and propose the Anchor-Guided Attack (AGA) to circumvent existing safeguards. Specifically, AGA aligns the protected model with this anchor to recover the transformation matrix analytically. Extensive evaluations validate that AGA consistently bypasses both individual and composite defenses under realistic defense-agnostic scenarios. Furthermore, we provide Anchor-Repulsive Fine-tuning (ARF), a defense method to mitigate the anchor dominance leveraged by AGA. Empirical results confirm that ARF effectively defeats the proposed attack. Our code is available at https://github.com/krumpguo/secure-merge-attack.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Systematically Analyzing Prompt Injection Vulnerabilities in Diverse LLM Architectures
This study systematically analyzes the vulnerability of 36 large language models (LLMs) to various prompt injection attacks, a technique that leverages carefully crafted prompts to elicit malicious LLM behavior. Across 1…
MisinformationRethinking IC layout vulnerability: Simulation-based hardware Trojan threat assessment with high fidelity
Due to the escalating complexity of chip design and the exorbitant cost of building cutting-edge manufacturing facilities, outsourcing the fabrication of Integrated Circuits (ICs) is prevalent in modern semiconductor ind…
Side Channel AnalysisExposing the Systematic Vulnerability of Open-Weight Models to Prefill Attacks
As the capabilities of large language models continue to advance, so does their potential for misuse. While closed-source models typically rely on external defenses, open-weight models must primarily depend on internal s…
Adversarial Robustness for Machine Learning Cyber Defenses Using Log Data
There has been considerable and growing interest in applying machine learning for cyber defenses. One promising approach has been to apply natural language processing techniques to analyze logs data for suspicious behavi…
Adversarial AttackAdversarial RobustnessBIG-bench Machine LearningDeep Reinforcement LearningEmoRAG: Evaluating RAG Robustness to Symbolic Perturbations
Retrieval-Augmented Generation (RAG) systems are increasingly central to robust AI, enhancing large language model (LLM) faithfulness by incorporating external knowledge. However, our study unveils a critical, overlooked…