---
format: "aidr-story-markdown/v1"
id: "df16ae68229af0edfaa1e537e6b7f6d3c040a95e31e23d46919c4a833158c8b6"
canonical_url: "https://aidr.today/df16ae68?lang=en"
title: "Interactive Backdoor Attacks in LLM-Based Proactive Dialogue Systems"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T05:00:00.000Z"
category: "Research"
topics: ["safety"]
source_urls: ["https://arxiv.org/abs/2611.06969"]
summary: "LLM-based proactive dialogue systems provide goal-directed support through sustained interaction. Fine-tuning these models on externally supplied dialogue data introduces backdoor risks. An interactive backdoor attack framework uses training-data poisoning to steer post-trigger, multi-turn interactions toward an attacker-specified objective."
---

# Interactive Backdoor Attacks in LLM\-Based Proactive Dialogue Systems

> [Open the canonical story](<https://aidr.today/df16ae68?lang=en>)

**Published:** 2026-11-06T05:00:00.000Z
**Category:** Research
**Topics:** safety

## Summary

LLM\-based proactive dialogue systems provide goal\-directed support through sustained interaction\. Fine\-tuning these models on externally supplied dialogue data introduces backdoor risks\. An interactive backdoor attack framework uses training\-data poisoning to steer post\-trigger, multi\-turn interactions toward an attacker\-specified objective\.

## Sources

- [Story source](<https://arxiv.org/abs/2611.06969>)