🧪 Test?View on arXiv
Interactive Backdoor Attacks in LLM-Based Proactive Dialogue Systems
Not provided
backdoor attacksdialogue systemsfine-tuning
2611.06969
Builder Relevance
3h ago70%
Abstract
This paper presents an interactive backdoor attack framework for LLM-based proactive dialogue systems that allows attackers to steer multi-turn interactions toward specific objectives using poisoned training data.
Reality Card
Core Claim
The study demonstrates that interactive backdoor attacks can effectively manipulate dialogue systems to achieve attacker-specified objectives while maintaining normal utility.
Method / Result
All three attack methods achieved their objectives in experiments, preserving normal utility.
Limitations
The tested training-data filters failed to retain nearly all poisoned dialogues, raising concerns about the effectiveness of defenses.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.
No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.