cs.CLNov 4, 2026

Interactive Backdoor Attacks in LLM-Based Proactive Dialogue Systems

Authors: Zhiyang Xia, Yuming Shang, Xi Zhang

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.

Figures & tables

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Oct 6, 2026

The Model Plants the Trigger: Answer-Side Backdoor Attacks in Multi-Turn Large Language Models

Safety alignment in Large Language Models (LLMs) remains vulnerable to backdoor attacks. Existing LLM backdoors are almost all input-centric: activation depends on explicit trigger patterns in the user input, so modern guardrails are built to sanitize the input space. We challenge this assumption with a novel answer-side backdoor for multi-turn dialogue. Instead of inserting the trigger into the input, the adversary uses a benign first-turn prompt to naturally induce the model to generate a specific, seemingly innocuous word. Once merged into the dialogue history, this self-generated word becomes the trigger. When a later harmful query arrives, the model detects its own trigger and bypasses its safety refusal, while the user input stays perfectly clean. Across four LLMs, our attack reaches near-perfect Attack Success Rates, approaching 100% at only a 5% poisoning rate, while preserving general utility and clean-input safety, and it evades mainstream input-centric defenses. Representation-level analysis shows that the self-generated trigger consistently suppresses the model's refusal signal, exposing a critical blind spot in current LLM defenses.
Apr 23, 2026

Stealthy Backdoor Attacks against LLMs Based on Natural Style Triggers

The growing application of large language models (LLMs) in safety-critical domains has raised urgent concerns about their security. Many recent studies have demonstrated the feasibility of backdoor attacks against LLMs. However, existing methods suffer from three key shortcomings: explicit trigger patterns that compromise naturalness, unreliable injection of attacker-specified payloads in long-form generation, and incompletely specified threat models that obscure how backdoors are delivered and activated in practice. To address these gaps, we present BadStyle, a complete backdoor attack framework and pipeline. BadStyle leverages an LLM as a poisoned sample generator to construct natural and stealthy poisoned samples that carry imperceptible style-level triggers while preserving semantics and fluency. To stabilize payload injection during fine-tuning, we design an auxiliary target loss that reinforces the attacker-specified target content in responses to poisoned inputs and penalizes its emergence in benign responses. We further ground the attack in a realistic threat model and systematically evaluate BadStyle under both prompt-induced and PEFT-based injection strategies. Extensive experiments across seven victim LLMs, including LLaMA, Phi, DeepSeek, and GPT series, demonstrate that BadStyle achieves high attack success rates (ASRs) while maintaining strong stealthiness. The proposed auxiliary target loss substantially improves the stability of backdoor activation, yielding an average ASR improvement of around 30% across style-level triggers. Even in downstream deployment scenarios unknown during injection, the implanted backdoor remains effective. Moreover, BadStyle consistently evades representative input-level defenses and bypasses output-level defenses through simple camouflage.
May 14, 2026

MetaBackdoor: Exploiting Positional Encoding as a Backdoor Attack Surface in LLMs

Backdoor attacks pose a serious security threat to large language models (LLMs), which are increasingly deployed as general-purpose assistants in safety- and privacy-critical applications. Existing LLM backdoors rely primarily on content-based triggers, requiring explicit modification of the input text. In this work, we show that this assumption is unnecessary and limiting. We introduce MetaBackdoor, a new class of backdoor attacks that exploits positional information as the trigger, without modifying textual content. Our key insight is that Transformer-based LLMs necessarily encode token positions to process ordered sequences. As a result, length-correlated positional structure is reflected in the model's internal computation and can be used as an effective non-content trigger signal. We demonstrate that even a simple length-based positional trigger is sufficient to activate stealthy backdoors. Unlike prior attacks, MetaBackdoor operates on visibly and semantically clean inputs and enables qualitatively new capabilities. We show that a backdoored LLM can be induced to disclose sensitive internal information, including proprietary system prompts, once a length condition is satisfied. We further demonstrate a self-activation scenario, where normal multi-turn interaction can move the conversation context into the trigger region and induce malicious tool-call behavior without attacker-supplied trigger text. In addition, MetaBackdoor is orthogonal to content-based backdoors and can be composed with them to create more precise and harder-to-detect activation conditions. Our results expand the threat model of LLM backdoors by revealing positional encoding as a previously overlooked attack surface. This challenges defenses that focus on detecting suspicious text and highlights the need for new defense strategies that explicitly account for positional triggers in modern LLM architectures.