mlsanityData to 14 Oct 2026

SafeTy Reasoning Elicitation Alignment for Multi-Turn Dialogues

arXiv 2506.00668 ↗First posted 31 May 2025cs.CLcited by 7

On the map: LLM reasoning → Reasoning for agents, planning and code

Abstract (arXiv)

Malicious attackers can exploit large language models (LLMs) by engaging them in multi-turn dialogues to achieve harmful objectives, posing significant safety risks to society. To address this challenge, we propose a novel defense mechanism: SafeTy Reasoning Elicitation Alignment for Multi-Turn Dialogues (STREAM). STREAM defends LLMs against multi-turn attacks while preserving their functional capabilities. Our approach involves constructing a human-annotated dataset, the Safety Reasoning Multi-turn Dialogues dataset, which is used to fine-tune a plug-and-play safety reasoning moderator. This model is designed to identify malicious intent hidden within multi-turn conversations and alert the target LLM of potential risks. We evaluate STREAM across multiple LLMs against prevalent multi-turn attack strategies. Experimental results demonstrate that our method significantly outperforms existing defense techniques, reducing the Attack Success Rate (ASR) by 51.2%, all while maintaining comparable LLM capability.

Key facts. Cited by 7 papers in the map (data through 14 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 18 of its 28 references that are in the map. First posted on arXiv 31 May 2025.

Builds on

18 of 28 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025H-CoT: Hijacking the Chain-of-Thought Safety Reasoning Mechanism to Jailbreak Large Reasoning Models, Including OpenAI o1/o3, DeepSeek-R1, and Gemini 2.0 Flash Thinking18 Feb
  2. 2024OpenAI o1 System Card21 Dec

Other references · 16

  1. 2024The Llama 3 Herd of Modelscited by 14,715
  2. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  3. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  4. 2024GPT-4o System Cardcited by 4,588
  5. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  6. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,154
  7. 2023GPT-4 Is Too Smart To Be Safe: Stealthy Chat with LLMs via Ciphercited by 423
  8. 2024Great, Now Write an Article About That: The Crescendo Multi-Turn LLM Jailbreak Attackcited by 390
  9. 2025Llama-Nemotron: Efficient Reasoning Modelscited by 158
  10. 2024LLM Defenses Are Not Robust to Multi-Turn Human Jailbreaks Yetcited by 151
  11. 2025X-Teaming: Multi-Turn Jailbreaks and Defenses with Adaptive Multi-Agentscited by 125
  12. 2024Weak-to-Strong Jailbreaking on Large Language Modelscited by 107
4 more
  1. 2024LLMs know their vulnerabilities: Uncover Safety Gaps through Natural Distribution Shiftscited by 86
  2. 2024Chain of Attack: a Semantic-Driven Contextual Multi-Turn attacker for LLMcited by 77
  3. 2023Examining LLMs' Uncertainty Expression Towards Questions Outside Parametric Knowledgecited by 32
  4. 2024Granite Guardiancited by 25

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 4; 2026: 3.

7 papers.

  1. 2025ScamAgents: How AI Agents Can Simulate Human-Level Scam Callscited by 4
  2. 2025AURA: Affordance-Understanding and Risk-aware Alignment Technique for Large Language Modelscited by 2
  3. 2025Rethinking Reasoning: A Survey on Reasoning-based Backdoors in LLMscited by 2
  4. 2025LLaVAShield: Safeguarding Multimodal Multi-Turn Dialogues in Vision-Language Modelscited by 1
  5. 2026JMedEthicBench: A Multi-Turn Conversational Benchmark for Evaluating Medical Safety in Japanese Large Language Modelscited by 1
  6. 2026Conversations Risk Detection LLMs in Financial Agents via Multi-Stage Generative Rolloutcited by 0
  7. 2026RoTRAG: Rule of Thumb Reasoning for Conversation Harm Detection with Retrieval-Augmented Generationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026One Turn Too Late: Learning When to Intervene Against Multi-Turn Malicious Intent
  2. 2024Speak Out of Turn: Safety Vulnerability of Large Language Models in Multi-turn Dialogue
  3. 2025MUSE: MCTS-Driven Red Teaming Framework for Enhanced Multi-Turn Dialogue Safety in Large Language Models
  4. 2025SafeMT: Multi-turn Safety for Multimodal Language Models
  5. 2025Reasoning-Augmented Conversation for Multi-Turn Jailbreak Attacks on Large Language Models
  6. 2025STAIR: Improving Safety Alignment with Introspective Reasoning
  7. 2026Beyond Unsafe Detection: Counterfactually Anchored Evidence Attribution for Multi-Turn LLM Safety Failures
  8. 2026Towards Multi-modal Multi-turn Safety: From Agentic Interaction to Strategic Alignment