Oriel — AI & technology worth noticing, every day.

Interactive Backdoor Attacks in LLM-Based Proactive Dialogue Systems

Paper · https://arxiv.org/abs/2611.06969

arXiv:2611.06969v1 Announce Type: new Abstract: LLM-based proactive dialogue systems provide goal-directed support through sustained interaction. Fine-tuning their underlying models on externally supplied dialogue data introduces backdoor risks. Existing LLM backdoor studies primarily focus on eliciting an attacker-specified response upon trigger activation. We present an interactive backdoor attack framework that uses training-data poisoning to steer post-trigger, multi-turn interactions toward an attacker-specified objective. Poisoned fine-tuning dialogues implant a trigger-conditioned guidance-policy bias that ordinary user utterances can activate, without deployment-time prompt modification or further attacker intervention. Specifically, we instantiate three complementary attacks under progressively expanded post-trigger interaction settings: (1) Direct Attack, which targets an immediate response deviation upon trigger activation, retaining the conventional trigger--response setting; (2) Constrained Progressive Attack, which progressively advances the attack objective along a predefined multi-turn trajectory; and (3) Unconstrained Progressive Attack, which extends target-directed steering to open-ended interactions with varied user continuations. Using psychological counseling as a representative setting, experiments on two LLM backbones and two dialogue datasets show that all three attacks effectively achieve their objectives under their respective interaction settings while largely preserving normal utility. Separate defense evaluations show that the tested training-data filters retain nearly all poisoned dialogues, while three post-admission baselines leave substantial residual attack-specific success.

First seen by Oriel on .

AI summary

The paper presents training-data poisoning attacks that backdoor proactive dialogue systems, steering multi-turn conversations toward attacker-defined objectives after ordinary user utterances activate a trigger-conditioned policy bias.

  • The framework requires neither deployment-time prompt modification nor further attacker intervention after poisoned fine-tuning dialogues implant the backdoor.
  • Three variants cover immediate response deviation, progression along a predefined multi-turn trajectory, and target-directed steering during open-ended interactions with varied user continuations.
  • Psychological counseling experiments across two LLM backbones and two dialogue datasets found all attacks effective while largely preserving normal utility.
  • Tested training-data filters retained nearly all poisoned dialogues, while three post-admission defenses left substantial residual attack-specific success.

#llm security #backdoor attacks #dialogue systems #data poisoning

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
arXiv AI Papersoai:arXiv.org:2611.06969v1https://arxiv.org/abs/2611.06969 Read discussion