mlsanityData to 7 Oct 2026

The dark deep side of DeepSeek: Fine-tuning attacks against the safety alignment of CoT-enabled models

arXiv 2502.01225 ↗First posted 3 Feb 2025cs.CR cs.AIcited by 12

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

also in 2 more
Abstract (arXiv)

Large language models are typically trained on vast amounts of data during the pre-training phase, which may include some potentially harmful information. Fine-tuning attacks can exploit this by prompting the model to reveal such behaviours, leading to the generation of harmful content. In this paper, we focus on investigating the performance of the Chain of Thought based reasoning model, DeepSeek, when subjected to fine-tuning attacks. Specifically, we explore how fine-tuning manipulates the model's output, exacerbating the harmfulness of its responses while examining the interaction between the Chain of Thought reasoning and adversarial inputs. Through this study, we aim to shed light on the vulnerability of Chain of Thought enabled models to fine-tuning attacks and the implications for their safety and ethical deployment.

Key facts. Cited by 12 papers in the map (data through 7 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 2 of its 12 references that are in the map. First posted on arXiv 3 Feb 2025.

Builds on

2 of 12 references.

References · 2

  1. 2023Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!cited by 1,106
  2. 2024Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMscited by 158

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 9; 2026: 3.

12 papers.

  1. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonablecited by 120
  2. 2025SafeMLRM: Demystifying Safety in Multi-modal Large Reasoning Modelscited by 29
  3. 2025Towards Understanding the Safety Boundaries of DeepSeek Models: Evaluation and Findingscited by 24
  4. 2025Large Language Model (LLM) for Software Security: Code Analysis, Malware Analysis, Reverse Engineeringcited by 15
  5. 2025UnsafeChain: Enhancing Reasoning Model Safety via Hard Casescited by 7
  6. 2025DeepSeek-Inspired Exploration of RL-based LLMs and Synergy with Wireless Networks: A Surveycited by 6
  7. 2025When Models Outthink Their Safety: Unveiling and Mitigating Self-Jailbreak in Large Reasoning Modelscited by 5
  8. 2025AdvChain: Adversarial Chain-of-Thought Tuning for Robust Safety Alignment of Large Reasoning Modelscited by 4
  9. 2026RouteHijack: Routing-Aware Attack on Mixture-of-Experts LLMscited by 4
  10. 2025ToxiFrench: Benchmarking and Enhancing Language Models via CoT Fine-Tuning for French Toxicity Detectioncited by 3
  11. 2026Surgery: Mitigating Harmful Fine-Tuning for Large Language Models via Attention Sinkcited by 3
  12. 2026AI-Generated PowerShell Malware: An Experimental Framework and Datasetcited by 0

Similar, not cited

Close in wording, no citation either way

  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 Thinking
  2. 2025Refining Input Guardrails: Enhancing LLM-as-a-Judge Efficiency Through Chain-of-Thought Fine-Tuning and Alignment
  3. 2025BadThink: Triggered Overthinking Attacks on Chain-of-Thought Reasoning in Large Language Models
  4. 2025The Hidden Risks of Large Reasoning Models: A Safety Assessment of R1
  5. 2025SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilities
  6. 2025Thought Purity: A Defense Framework For Chain-of-Thought Attack
  7. 2024No Two Devils Alike: Unveiling Distinct Mechanisms of Fine-tuning Attacks
  8. 2026The Fragility of Chain-of-Thought Monitoring Across Typologically Diverse Languages