# First Token Matters: Understanding Safety Collapse in Large Reasoning Models

arXiv: https://arxiv.org/abs/2609.18471 · HTML: https://mlsanity.com/paper/2609.18471/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning.
- Builds on 26 of its 46 references that are in the map.
- First posted on arXiv 16 Sep 2026.

## Placed in

- LLM reasoning / [Reasoning for agents, planning and code](https://mlsanity.com/lr/reasoning-for-agents-planning-and-code/)
- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)
- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 26 of 26)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://mlsanity.com/paper/2501.12948/) (2025)
- [Let's Verify Step by Step](https://mlsanity.com/paper/2305.20050/) (2023)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Qwen2 Technical Report](https://mlsanity.com/paper/2407.10671/) (2024)
- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [OpenAI o1 System Card](https://mlsanity.com/paper/2412.16720/) (2024)
- [XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models](https://mlsanity.com/paper/2308.01263/) (2023)
- [A StrongREJECT for Empty Jailbreaks](https://mlsanity.com/paper/2402.10260/) (2024)
- [WildTeaming at Scale: From In-the-Wild Jailbreaks to (Adversarially) Safer Language Models](https://mlsanity.com/paper/2406.18510/) (2024)
- [H-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 Thinking](https://mlsanity.com/paper/2502.12893/) (2025)
- [Steering Language Model Refusal with Sparse Autoencoders](https://mlsanity.com/paper/2411.11296/) (2024)
- [STAR-1: Safer Alignment of Reasoning LLMs with 1K Data](https://mlsanity.com/paper/2504.01903/) (2025)
- [Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons](https://mlsanity.com/paper/2406.14144/) (2024)
- [SafeSwitch: Steering Unsafe LLM Behavior via Internal Activation Signals](https://mlsanity.com/paper/2502.01042/) (2025)
- [SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment](https://mlsanity.com/paper/2505.14667/) (2025)
- [PromptGuard: Soft Prompt-Guided Unsafe Content Moderation for Text-to-Image Models](https://mlsanity.com/paper/2501.03544/) (2025)
- [Chain-of-Thought Hijacking](https://mlsanity.com/paper/2510.26418/) (2025)
- [Large Reasoning Models Learn Better Alignment from Flawed Thinking](https://mlsanity.com/paper/2510.00938/) (2025)
- [Self-Jailbreaking: Language Models Can Reason Themselves Out of Safety Alignment After Benign Reasoning Training](https://mlsanity.com/paper/2510.20956/) (2025)
- [Refusal Falls off a Cliff: How Safety Alignment Fails in Reasoning?](https://mlsanity.com/paper/2510.06036/) (2025)
- [Universal Jailbreak Suffixes Are Strong Attention Hijackers](https://mlsanity.com/paper/2506.12880/) (2025)
- [Safety Alignment of Large Language Models via Contrasting Safe and Harmful Distributions](https://mlsanity.com/paper/2406.16743/) (2024)
- [Read the Scene, Not the Script: Outcome-Aware Safety for LLMs](https://mlsanity.com/paper/2510.04320/) (2025)
- [When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Models](https://mlsanity.com/paper/2510.21285/) (2025)
- [Safety Recovery in Reasoning Models Is Only a Few Early Steering Steps Away](https://mlsanity.com/paper/2602.11096/) (2026)

## Abstract (arXiv)

Large Reasoning Models (LRMs) exhibit strong problem-solving abilities, yet their safety alignment often degrades when handling harmful queries. Existing approaches to improving safety largely rely on additional training or preference optimization, while offering limited understanding of the internal mechanisms behind safety failures. In this work, we investigate this failure through a token-level positional analysis of refusal dynamics and identify a localized vulnerability at the onset of reasoning, which we term Onset Refusal Collapse (ORC). We find that the refusal-related signal of LRMs drops sharply at the first generated token under harmful queries, which is associated with unsafe response generation. Motivated by this finding, we propose SafeToken, a lightweight inference-time intervention that injects a learned continuous safety anchor precisely at reasoning onset. Despite updating only a single token embedding, SafeToken effectively mitigates ORC, improves safety on harmful-query benchmarks, and largely preserves reasoning utility. These results suggest that safety failures in LRMs can arise from a transient breakdown at the critical transition from understanding to generation.