aiqbench analysis
This research exposes a critical vulnerability in task vector arithmetic where asymmetric backdoors can persist across learning, forgetting, and analogy operations with near-perfect success rates. Current defense mechanisms fail to detect or mitigate these threats, highlighting the inherent security risks of modular model adaptation techniques used in large-scale pre-trained systems. The findings underscore an urgent need for robust countermeasures before task vector methods are widely deployed in production environments.
BADTV: Unveiling Backdoor Threats in Third-Party Task Vectors
arXiv:2501.02373v3 Announce Type: replace Abstract: Task arithmetic in large-scale pre-trained models enables agile adaptation to diverse downstream tasks without extensive retraining. By leveraging task vectors (TVs), users can perform modular updates through simple arithmetic operations like addition and subtraction. Yet, this flexibility presents new security challenges. In this paper, we investigate how TVs are vulnerable to backdoor attacks, revealing how malicious actors can exploit them to compromise model integrity. By creating composite backdoors that are designed asymmetrically, we introduce BadTV, a backdoor attack specifically crafted to remain effective simultaneously under task learning, forgetting, and analogy operations. Extensive experiments show that BadTV achieves near-perfect attack success rates across diverse scenarios, posing a serious threat to models relying on task arithmetic. We also evaluate current defenses, finding they fail to detect or mitigate BadTV. Our results highlight the urgent need for robust countermeasures to secure TVs in real-world deployments.
Abstract:Task arithmetic in large-scale pre-trained models enables agile adaptation to diverse downstream tasks without extensive retraining. By leveraging task vectors (TVs), users can perform modular updates through simple arithmetic operations like addition and subtraction. Yet, this flexibility presents new security challenges. In this paper, we investigate how TVs are vulnerable to backdoor attacks, revealing how malicious actors can exploit them to compromise model integrity. By creating composite backdoors that are designed asymmetrically, we introduce BadTV, a backdoor attack specifically crafted to remain effective simultaneously under task learning, forgetting, and analogy operations. Extensive experiments show that BadTV achieves near-perfect attack success rates across diverse scenarios, posing a serious threat to models relying on task arithmetic. We also evaluate current defenses, finding they fail to detect or mitigate BadTV. Our results highlight the urgent need for robust countermeasures to secure TVs in real-world deployments.
Machine Learning (cs.LG); Cryptography and Security (cs.CR)
arXiv:2501.02373 [cs.LG]
(or arXiv:2501.02373v3 [cs.LG] for this version)
Full text under the publisher's open licence · 157 words
Related coverage
RogueMerge: Robust and Unified Attacks against LLM Model Merging
arXiv · Jun 3, 2026
RogueMerge: Robust and Unified Attacks against LLM Model Merging
arXiv · Oct 14, 2026
Model Poisoning Against Federated Model Adaptation with Chain of Bit-Flips
arXiv · Jun 9, 2026
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · Aug 3, 2026