One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue

TL;DR

This paper introduces a method to detect the earliest turn in multi-turn dialogues that enables harmful actions, using a new dataset and a turn-level monitor to improve safety in large language models.

Contribution

The authors propose TurnGate, a turn-level monitor for harmful intent detection, supported by the Multi-Turn Intent Dataset (MTID), enhancing safety measures against distributed malicious prompts.

Findings

TurnGate outperforms existing baselines in harmful intent detection.

MTID enables effective training and evaluation of turn-level safety monitors.

TurnGate generalizes well across different domains and attacker pipelines.

Abstract

Hidden malicious intent in multi-turn dialogue poses a growing threat to deployed large language models (LLMs). Rather than exposing a harmful objective in a single prompt, increasingly capable attackers can distribute their intent across multiple benign-looking turns. Recent studies show that even modern commercial models with advanced guardrails remain vulnerable to such attacks despite advances in safety alignment and external guardrails. In this work, we address this challenge by detecting the earliest turn at which delivering the candidate response would make the accumulated interaction sufficient to enable harmful action. This objective requires precise turn-level intervention that identifies the harm-enabling closure point while avoiding premature refusal of benign exploratory conversations. To further support training and evaluation, we construct the Multi-Turn Intent Dataset (MTID), which contains branching attack rollouts, matched benign hard negatives, and annotations of the earliest harm-enabling turns. We show that MTID helps enable a turn-level monitor TurnGate, which substantially outperforms existing baselines in harmful-intent detection while maintaining low over-refusal rates. TurnGate further generalizes across domains, attacker pipelines, and target models. Our code is available at https://github.com/Graph-COM/TurnGate.