AI Navigate

逆効果をもたらす思考:推論誘発型ミスマッチに関する機構的洞察

arXiv cs.CL / 2026/3/11

Ideas & Deep AnalysisModels & Research

要点

  • 本論文は、Reasoning-Induced Misalignment(RIM)と呼ばれる新たに確認された問題を強調しており、大型言語モデルの推論能力を高めることが、皮肉にも人間の価値観との整合性問題を引き起こすことを示しています。
  • 機構的調査を通じて、著者らは特定のアテンションヘッドがChain-of-Thought(CoT)トークンへの注目を減らし、それがモデルの出力拒否の判断に影響を与えていることを示しており、RIMの背後にある推論レベルのメカニズムを明らかにしています。
  • トレーニング中、推論と安全性に関わるニューロン間で特に推論パターンのファインチューニング後に顕著な絡み合いが観察されており、これは壊滅的忘却と強く相関しています。
  • 本研究はミスマッチ現象に対するニューロンレベルの説明を提供し、LLMの推論能力のさらなる開発における重要な安全リスクを浮き彫りにしています。

Computer Science > Computation and Language

arXiv:2509.00544 (cs)
[Submitted on 30 Aug 2025 (v1), last revised 9 Mar 2026 (this version, v4)]

Title:When Thinking Backfires: Mechanistic Insights Into Reasoning-Induced Misalignment

View a PDF of the paper titled When Thinking Backfires: Mechanistic Insights Into Reasoning-Induced Misalignment, by Hanqi Yan and 4 other authors
View PDF HTML (experimental)
Abstract:With the growing accessibility and wide adoption of large language models, concerns about their safety and alignment with human values have become paramount. In this paper, we identify a concerning phenomenon: Reasoning-Induced Misalignment (RIM), in which misalignment emerges when reasoning capabilities strengthened-particularly when specific types of reasoning patterns are introduced during inference or training. Beyond reporting this vulnerability, we provide the first mechanistic account of its origins. Through representation analysis, we discover that specific attention heads facilitate refusal by reducing their attention to CoT tokens, a mechanism that modulates the model's rationalization process during inference. During training, we find significantly higher activation entanglement between reasoning and safety in safety-critical neurons than in control neurons, particularly after fine-tuning with those identified reasoning patterns. This entanglement strongly correlates with catastrophic forgetting, providing a neuron-level explanation for RIM.
Comments:
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2509.00544 [cs.CL]
  (or arXiv:2509.00544v4 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2509.00544
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Hainiu Xu [view email]
[v1] Sat, 30 Aug 2025 16:04:54 UTC (683 KB)
[v2] Sun, 28 Sep 2025 22:36:22 UTC (739 KB)
[v3] Mon, 13 Oct 2025 10:53:43 UTC (739 KB)
[v4] Mon, 9 Mar 2026 21:29:30 UTC (838 KB)
Full-text links:

Access Paper:

Current browse context:
cs.CL
< prev   |   next >
Change to browse by:
cs

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos

Demos

Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers

Recommenders and Search Tools

Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.