RogueMerge: Robust and Unified Attacks against LLM Model Merging
cs.CR, cs.LG
Submitted: 2026-06-02
Updated: 2026-10-12
Comments: To appear in IEEE Symposium on Security and Privacy (S&P), 2027
Code: https://github.com/bigcode-project/bigcode-evaluation-harness
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Editing Models with Task Arithmetic
- MergeBench: A Benchmark for Merging Domain-Specialized LLMs
- The Llama 3 Herd of Models
- Qwen Technical Report
- Gemma 2: Improving Open Language Models at a Practical Size
- Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities
- LoBAM: LoRA-Based Backdoor Attack on Model Merging
- Merge Hijacking: Backdoor Attacks to Model Merging of Large Language Models
- BackdoorLLM: A Comprehensive Benchmark for Backdoor Attacks and Defenses on Large Language Models
- Universal and Transferable Adversarial Attacks on Aligned Language Models
- Effective Prompt Extraction from Language Models
- Explaining and Harnessing Adversarial Examples
- Tulu 3: Pushing Frontiers in Open Language Model Post-Training
- Magicoder: Empowering Code Generation with OSS-Instruct
- LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models
- DELLA-Merging: Reducing Interference in Model Merging through Magnitude-Based Sampling
- Activation-Informed Merging of Large Language Models
- Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs
- Can You Really Backdoor Federated Learning?
- Fine-Tuning Is All You Need to Mitigate Backdoor Attacks
Related papers
- SoK: AI-Augmented Binary Reversing
- Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
- Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
- Efficient Fuzzy PSI under One-Sided Assumptions
- Sealing the Audit-Runtime Gap for LLM Skills
- Token Composition: A Graph Based on EVM Logs