{"slug": "first-token-matters-understanding-safety-collapse-in-large-reasoning-models", "title": "First Token Matters: Understanding Safety Collapse in Large Reasoning Models", "summary": "A new arXiv paper (2609.18471v1) identifies a failure mode it calls Onset Refusal Collapse (ORC), in which the refusal-related signal of Large Reasoning Models drops sharply at the first generated token under harmful queries, correlating with unsafe response generation. The authors propose SafeToken, a lightweight inference-time intervention that injects a learned continuous safety anchor at reasoning onset; updating only a single token embedding, SafeToken mitigates ORC, improves safety on harmful-query benchmarks, and largely preserves reasoning utility. The work attributes safety failures in LRMs to a transient breakdown at the transition from understanding to generation rather than to a lack of training.", "body_md": "arXiv:2609.18471v1 Announce Type: new \nAbstract: 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.", "url": "https://wpnews.pro/news/first-token-matters-understanding-safety-collapse-in-large-reasoning-models", "canonical_source": "https://www.machinebrief.com/news/first-token-matters-understanding-safety-collapse-in-large-r-7es1", "published_at": "2026-09-17 04:00:00+00:00", "updated_at": "2026-09-17 05:54:46.518347+00:00", "lang": "en", "topics": ["ai-safety", "large-language-models", "ai-research", "artificial-intelligence"], "entities": ["Large Reasoning Models", "Onset Refusal Collapse", "SafeToken", "arXiv"], "alternates": {"html": "https://wpnews.pro/news/first-token-matters-understanding-safety-collapse-in-large-reasoning-models", "markdown": "https://wpnews.pro/news/first-token-matters-understanding-safety-collapse-in-large-reasoning-models.md", "text": "https://wpnews.pro/news/first-token-matters-understanding-safety-collapse-in-large-reasoning-models.txt", "jsonld": "https://wpnews.pro/news/first-token-matters-understanding-safety-collapse-in-large-reasoning-models.jsonld"}}