社会层面 · safety论文趋势 2023–2026
AI 社会议题：safety
统计范围：NeurIPS、ICLR、ICML、CVPR 四个会议，2023–2026 年，共 688 篇被归入该主题的论文。
主要结论
- 社会层面 · safety论文从 2023 年的 4 篇增长到 2026 年的 370 篇，三年间扩大到约 93 倍。
- ICML 是发社会层面 · safety论文最多的会议，2023–2026 年共 364 篇，占四个会议总量的 53%。
- 交叉方向里涨得最快的是对齐与安全：同时属于两个主题的论文从 2023 年的 0 篇增至 2026 年的 137 篇。
- 在「对齐与负责任 AI」的 5 个主题里，它的 2026 年论文数排第 3；同组增长最快的社会层面 · privacy是 6.6 倍，社会层面 · safety是 93 倍。
各会议论文数
按年份和会议统计
| 年份 | NeurIPS | ICLR | ICML | CVPR | 合计 |
|---|---|---|---|---|---|
| 2023 | 2 | 0 | 2 | 0 | 4 |
| 2024 | 23 | 11 | 6 | 1 | 41 |
| 2025 | 131 | 11 | 130 | 1 | 273 |
| 2026 | 121 | 20 | 226 | 3 | 370 |
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
- NeurIPS 2026BarrierSteer: LLM Safety via Learning Barrier Steering
- ICML 2026Anti-Backdoor Coreset Selection via Cumulative Entropy
- ICLR 2026Bi-Criteria Metric Distortion
- CVPR 2026Omni-MMSI Toward Identity-attributed Social Interaction Understanding
- NeurIPS 2025Safety Pretraining: Toward the Next Generation of Safe AI
- ICML 2025$S^2$FGL: Spatial Spectral Federated Graph Learning
- ICLR 2025The Hidden Cost of Waiting for Accurate Predictions
- CVPR 2025Do We Really Need Curated Malicious Data for Safety Alignment in Multi-modal Large Language Models?
- NeurIPS 2024Quantifying the Bitter Lesson: How Safety Benchmarks Measure Capabilities Instead of Safety
- ICML 2024Optimally Improving Cooperative Learning in a Social Setting
- ICLR 2024The Devil is in the Neurons: Interpreting and Mitigating Social Biases in Language Models
- CVPR 2024Digital Life Project: Autonomous 3D Characters with Social Intelligence
- NeurIPS 2023Strategyproof Voting under Correlated Beliefs
- ICML 2023Adaptive Coordination in Social Embodied Rearrangement
- NeurIPS 2026Beyond Feature Disruption: Boundary-Diverting Unlearnable Examples against Linear Probing
- ICML 2026AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation
- ICLR 2026PRISON: Unmasking the Criminal Potential of Large Language Models
- CVPR 2026Spot The Ball A Benchmark for Visual Social Inference
- NeurIPS 2025SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identification
- ICML 2025Adversaries Can Misuse Combinations of Safe Models
全部社会层面 · safety论文
共 688 篇，第 6 / 14 页
- NeurIPS 2026Agent MechSuits : Mechanistic Subspace Safety Steering for Multi-Turn CLI Agents
- NeurIPS 2026Agentic-imodels: Evolving agentic interpretability tools via autoresearch
- NeurIPS 2026AI-Mediated Communication Can Steer Collective Opinion
- NeurIPS 2026Amortized Linear-time Exact Shapley Value for Product-Kernel Methods
- NeurIPS 2026Are Well-Trained Surrogates Optimal? Rethinking the Surrogate Role with Instability for Unlearnable Examples
- NeurIPS 2026Attack Selection In Agentic AI Control Evaluations Meaningfully Decreases Safety
- NeurIPS 2026Auditor-Assisted Summary-Channel Verification for Hosted LLM Identity Substitution
- NeurIPS 2026BarrierSteer: LLM Safety via Learning Barrier Steering
- NeurIPS 2026Beyond Confidence: Rethinking Self-Assessments for Performance Prediction in LLMs
- NeurIPS 2026Beyond Feature Disruption: Boundary-Diverting Unlearnable Examples against Linear Probing
- NeurIPS 2026Beyond the Prompt: Leveraging Pre-Decoding States for Jailbreak Detection in dLLMs
- NeurIPS 2026Black-Box Inference of LLM Architectural Properties with Restrictive API Access
- NeurIPS 2026BSO: Safety Alignment Is Density Ratio Matching
- NeurIPS 2026CalArena: A Large Scale Post-Hoc Calibration Benchmark
- NeurIPS 2026CaMeLs Can Use Computers Too: System-level Security for Computer Use Agents
- NeurIPS 2026Can VLMs Reason When to Stop for Human Safety?
- NeurIPS 2026Castle-in-the-Air: Probing the Foundational Visual Deficits of MLLMs via Bottom-Up Cognitive Factors
- NeurIPS 2026Chatter Attack: Resource Consumption Attack for Large Language Models
- NeurIPS 2026Cognitive bias benchmarks should incorporate insights from ecological rationality
- NeurIPS 2026Community-Centered AI is Feasible and Beneficial for Impacted Communities
- NeurIPS 2026Compact Representations of Impact-Based Fair-Ranking Policies
- NeurIPS 2026Compressing Collections of Trees with Decision Equivalence
- NeurIPS 2026Conceal, Reconstruct, Jailbreak: Exploiting the Reconstruction--Concealment Tradeoff in MLLMs
- NeurIPS 2026DACE: Diversity-Driven Adversarial Co-Evolution for Robust LLM Safety Alignment
- NeurIPS 2026Decentralized AI Governance Must Decouple Policy Processing from Capability Enforcement
- NeurIPS 2026Decomposing One Professional-Framing Pipeline: Which Components Shift LLM Safety Boundaries?
- NeurIPS 2026Decoupled Safety Control: A Safety-Control Algorithm for Training-Free Safety Guidance
- NeurIPS 2026DIAL: A Bounded-Monotone Adapter with Closed-Form Residual $L_\infty$ Bounds for Scalar-Controlled Refusal Modulation
- NeurIPS 2026Dialectics of Alignment: Harnessing Unsafe Knowledge for Dynamic Safety Routing
- NeurIPS 2026Distributionally Robust Algorithmic Recourse for Tree Ensembles
- NeurIPS 2026Do Thinking Tokens Help with Safety?
- NeurIPS 2026Dual-Contrastive Sparse Autoencoders Reveal Features of Musical Interpretation
- NeurIPS 2026Dyna-Style Safety Augmented Reinforcement Learning: Staying Safe in the Face of Uncertainty
- NeurIPS 2026Embeddings for Preferences, Not Semantics
- NeurIPS 2026Escape the Context Manifold: Preventing DiT In-Context Editing via Conditional Flow Hijacking
- NeurIPS 2026Evaluation Awareness in Language Models Has Limited Effect on Behaviour
- NeurIPS 2026Expected Harm: Rethinking Safety Evaluation of (Mis)Aligned LLMs
- NeurIPS 2026Finding Interpretable Prompt-Specific Circuits in Language Models
- NeurIPS 2026Fine-tuning Does Not Reach All: Uneven Safety and Knowledge Dynamics in Language Models
- NeurIPS 2026From Post-Hoc to Ante-Hoc: Consistently Explainable Semi-Supervised Time Series Classification
- NeurIPS 2026From Tokens to Tactics: Adversarial Text Optimization in an Axis-Aligned Rhetorical Strategy Space for Harmful Content Detection
- NeurIPS 2026GeoBiaset: A Counterfactual Benchmark for Demographic Bias in World-Level Geolocalization
- NeurIPS 2026Gradient-based Graph Structure Optimisation for Research Networks
- NeurIPS 2026Gradient-Mine Units: Scorched-Earth Strategy for Model Protection against Unauthorized Fine-Tuning
- NeurIPS 2026GradShield: Alignment Preserving Finetuning
- NeurIPS 2026Harnessing Textual Refusal Directions for Multimodal Safety
- NeurIPS 2026Hierarchical Conformal Classification
- NeurIPS 2026How Do Agentic LLMs Decide to Call Tools? A Scaffold Default Controlled by Suppression
- NeurIPS 2026How Hard is it to Rig a Benchmark? A Social Choice Analysis of Leaderboard Robustness
- NeurIPS 2026Human-AI Teaming Through the Lens of Calibration
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 查社会层面 · safety在 ICML 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": "social-safety"}
常见问题
- 2023–2026 年四大会议一共有多少篇社会层面 · safety论文？
- 共 688 篇，按年份分别是 2023 年 4 篇、2024 年 41 篇、2025 年 273 篇、2026 年 370 篇。
- NeurIPS 2026 有多少篇社会层面 · safety论文？
- NeurIPS 2026 有 121 篇社会层面 · safety论文，上一届是 131 篇。
- ICLR 2026 有多少篇社会层面 · safety论文？
- ICLR 2026 有 20 篇社会层面 · safety论文，上一届是 11 篇。
- ICML 2026 有多少篇社会层面 · safety论文？
- ICML 2026 有 226 篇社会层面 · safety论文，上一届是 130 篇。
- CVPR 2026 有多少篇社会层面 · safety论文？
- CVPR 2026 有 3 篇社会层面 · safety论文，上一届是 1 篇。
- 哪个会议的社会层面 · safety论文最多？
- ICML，2023–2026 年共 364 篇。