对齐与安全论文趋势 2023–2026

RLHF、偏好优化、红队与安全对齐

统计范围：NeurIPS、ICLR、ICML、CVPR 四个会议，2023–2026 年，共 4,558 篇被归入该主题的论文。

主要结论

每年论文数

各会议论文数

按年份和会议统计

对齐与安全论文数，按年份和会议
年份NeurIPSICLRICMLCVPR合计
202313887101112438
2024348156246125875
20254554733001351,363
20264806235512281,882

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

全部对齐与安全论文

年份全部2023202420252026
会议全部NeurIPSICLRICMLCVPR

共 300 篇，第 6 / 6 页

  1. ICML 2025Sparse Spectral Training and Inference on Euclidean and Hyperbolic Neural Networks
  2. ICML 2025Speak Easy: Eliciting Harmful Jailbreaks from LLMs with Simple Interactions
  3. ICML 2025SPRI: Aligning Large Language Models with Context-Situated Principles
  4. ICML 2025Square$\chi$PO: Differentially Private and Robust $\chi^2$-Preference Optimization in Offline Direct Alignment
  5. ICML 2025STAIR: Improving Safety Alignment with Introspective Reasoning
  6. ICML 2025Statistical and Computational Guarantees of Kernel Max-Sliced Wasserstein Distances
  7. ICML 2025Statistical Hypothesis Testing for Auditing Robustness in Language Models
  8. ICML 2025Strong and Weak Identifiability of Optimization-based Causal Discovery in Non-linear Additive Noise Models
  9. ICML 2025SWAN: SGD with Normalization and Whitening Enables Stateless LLM Training
  10. ICML 2025Symmetric Reinforcement Learning Loss for Robust Learning on Diverse Tasks and Model Scales
  11. ICML 2025Synthesizing Images on Perceptual Boundaries of ANNs for Uncovering and Manipulating Human Perceptual Variability
  12. ICML 2025TANGO: Clustering with Typicality-Aware Nonlocal Mode-Seeking and Graph-Cut Optimization
  13. ICML 2025Test-time Correlation Alignment
  14. ICML 2025Test-Time Preference Optimization: On-the-Fly Alignment via Iterative Textual Feedback
  15. ICML 2025TGDPO: Harnessing Token-Level Reward Guidance for Enhancing Direct Preference Optimization
  16. ICML 2025The Energy Loss Phenomenon in RLHF: A New Perspective on Mitigating Reward Hacking
  17. ICML 2025The Geometry of Refusal in Large Language Models: Concept Cones and Representational Independence
  18. ICML 2025The Hidden Dimensions of LLM Alignment: A Multi-Dimensional Analysis of Orthogonal Safety Directions
  19. ICML 2025The Jailbreak Tax: How Useful are Your Jailbreak Outputs?
  20. ICML 2025The Limits of Predicting Agents from Behaviour
  21. ICML 2025The Lock-in Hypothesis: Stagnation by Algorithm
  22. ICML 2025The Logical Implication Steering Method for Conditional Interventions on Transformer Generation
  23. ICML 2025The Missing Alignment Link of In-context Learning on Sequences
  24. ICML 2025The Perils of Optimizing Learned Reward Functions: Low Training Error Does Not Guarantee Low Regret
  25. ICML 2025The Relationship Between No-Regret Learning and Online Conformal Prediction
  26. ICML 2025Theoretical guarantees on the best-of-n alignment policy
  27. ICML 2025Thinking LLMs: General Instruction Following with Thought Generation
  28. ICML 2025TimePoint: Accelerated Time Series Alignment via Self-Supervised Keypoint and Descriptor Learning
  29. ICML 2025Tokenized Bandit for LLM Decoding and Alignment
  30. ICML 2025Topological Signatures of Adversaries in Multimodal Alignments
  31. ICML 2025Towards an Explainable Comparison and Alignment of Feature Embeddings
  32. ICML 2025Towards Cost-Effective Reward Guided Text Generation
  33. ICML 2025Towards Memorization Estimation: Fast, Formal and Free
  34. ICML 2025Towards Universal Offline Black-Box Optimization via Learning Language Model Embeddings
  35. ICML 2025Training Deep Learning Models with Norm-Constrained LMOs
  36. ICML 2025Training Flexible Models of Genetic Variant Effects from Functional Annotations using Accelerated Linear Algebra
  37. ICML 2025Transformative or Conservative? Conservation laws for ResNets and Transformers
  38. ICML 2025TRUST-VLM: Thorough Red-Teaming for Uncovering Safety Threats in Vision-Language Models
  39. ICML 2025TuCo: Measuring the Contribution of Fine-Tuning to Individual Responses of LLMs
  40. ICML 2025UDora: A Unified Red Teaming Framework against LLM Agents by Dynamically Hijacking Their Own Reasoning
  41. ICML 2025Unconstrained Robust Online Convex Optimization
  42. ICML 2025Understanding High-Dimensional Bayesian Optimization
  43. ICML 2025Understanding Input Selectivity in Mamba: Impact on Approximation Power, Memorization, and Associative Recall Capacity
  44. ICML 2025Understanding the Emergence of Multimodal Representation Alignment
  45. ICML 2025Understanding the Logic of Direct Preference Alignment through Logic
  46. ICML 2025Vulnerability-Aware Alignment: Mitigating Uneven Forgetting in Harmful Fine-Tuning
  47. ICML 2025WATCH: Adaptive Monitoring for AI Deployments via Weighted-Conformal Martingales
  48. ICML 2025Weak-to-Strong Jailbreaking on Large Language Models
  49. ICML 2025When Can Proxies Improve the Sample Complexity of Preference Learning?
  50. ICML 2025When, Where and Why to Average Weights?

在 Claude、Cursor 里直接问这个趋势

接上 Paperlist MCP 后，写综述、挑方向或准备投稿时，不用离开对话框就能拿到这页的数字：每年各会议的论文数、增长倍数、涨得最快的子方向和代表论文，AI 会按你的问题整理。

可以直接复制这句话问 AI

用 paperlist 查对齐与安全在 NeurIPS 2023–2026 的论文趋势，哪一年涨得最快？

AI 会调用的工具：get_topic_stats {"topic": "alignment-and-safety"}

两分钟接入 Paperlist MCP（Claude Code、Cursor、Codex、ChatGPT）

常见问题

2023–2026 年四大会议一共有多少篇对齐与安全论文？
共 4,558 篇，按年份分别是 2023 年 438 篇、2024 年 875 篇、2025 年 1,363 篇、2026 年 1,882 篇。
NeurIPS 2026 有多少篇对齐与安全论文？
NeurIPS 2026 有 480 篇对齐与安全论文，上一届是 455 篇。
ICLR 2026 有多少篇对齐与安全论文？
ICLR 2026 有 623 篇对齐与安全论文，上一届是 473 篇。
ICML 2026 有多少篇对齐与安全论文？
ICML 2026 有 551 篇对齐与安全论文，上一届是 300 篇。
CVPR 2026 有多少篇对齐与安全论文？
CVPR 2026 有 228 篇对齐与安全论文，上一届是 135 篇。
哪个会议的对齐与安全论文最多？
NeurIPS，2023–2026 年共 1,421 篇。

「对齐与负责任 AI」下的其他主题