社会层面 · safety论文趋势 2023–2026

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。

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 364 篇，第 6 / 8 页

251. ICML 2025Detecting Strategic Deception with Linear Probes
252. ICML 2025DIS-CO: Discovering Copyrighted Content in VLMs Training Data
253. ICML 2025Discovering Spoofing Attempts on Language Model Watermarks
254. ICML 2025Disparate Conditional Prediction in Multiclass Classifiers
255. ICML 2025Do Not Mimic My Voice : Speaker Identity Unlearning for Zero-Shot Text-to-Speech
256. ICML 2025Efficient and Privacy-Preserving Soft Prompt Transfer for LLMs
257. ICML 2025ELITE: Enhanced Language-Image Toxicity Evaluation for Safety
258. ICML 2025Embedding Safety into RL: A New Take on Trust Region Methods
259. ICML 2025Emoji Attack: Enhancing Jailbreak Attacks Against Judge LLM Detection
260. ICML 2025EraseAnything: Enabling Concept Erasure in Rectified Flow Transformers
261. ICML 2025Evaluating Neuron Explanations: A Unified Framework with Sanity Checks
262. ICML 2025Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanations
263. ICML 2025ExpProof : Operationalizing Explanations for Confidential Models with ZKPs
264. ICML 2025Extractive Structures Learned in Pretraining Enable Generalization on Finetuned Facts
265. ICML 2025Fair Clustering via Alignment
266. ICML 2025FairICP: Encouraging Equalized Odds via Inverse Conditional Permutation
267. ICML 2025Fast Estimation of Partial Dependence Functions using Trees
268. ICML 2025Flow Matching for Denoised Social Recommendation
269. ICML 2025From Pixels to Perception: Interpretable Predictions via Instance-wise Grouped Feature Selection
270. ICML 2025Gandalf the Red: Adaptive Security for LLMs
271. ICML 2025GaussMarker: Robust Dual-Domain Watermark for Diffusion Models
272. ICML 2025Graph Inverse Style Transfer for Counterfactual Explainability
273. ICML 2025How Do Transformers Learn Variable Binding in Symbolic Programs?
274. ICML 2025Improving LLM Safety Alignment with Dual-Objective Optimization
275. ICML 2025Improving Out-of-Distribution Detection via Dynamic Covariance Calibration
276. ICML 2025Improving Out-of-Distribution Detection with Markov Logic Networks
277. ICML 2025Inducing, Detecting and Characterising Neural Modules: A Pipeline for Functional Interpretability in Reinforcement Learning
278. ICML 2025InfoCons: Identifying Interpretable Critical Concepts in Point Clouds via Information Theory
279. ICML 2025Internal Causal Mechanisms Robustly Predict Language Model Out-of-Distribution Behaviors
280. ICML 2025Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations
281. ICML 2025Just Enough Shifts: Mitigating Over-Refusal in Aligned Language Models with Targeted Representation Fine-Tuning
282. ICML 2025Learning Classifiers That Induce Markets
283. ICML 2025Learning Safety Constraints for Large Language Models
284. ICML 2025Learning to Route LLMs with Confidence Tokens
285. ICML 2025Leveraging Predictive Equivalence in Decision Trees
286. ICML 2025LIMEFLDL: A Local Interpretable Model-Agnostic Explanations Approach for Label Distribution Learning
287. ICML 2025LlavaGuard: An Open VLM-based Framework for Safeguarding Vision Datasets and Models
288. ICML 2025MIB: A Mechanistic Interpretability Benchmark
289. ICML 2025Mind the Gap: A Practical Attack on GGUF Quantization
290. ICML 2025Minimalist Concept Erasure in Generative Models
291. ICML 2025Mixture of Experts Made Intrinsically Interpretable
292. ICML 2025MP-Nav: Enhancing Data Poisoning Attacks against Multimodal Learning
293. ICML 2025MTL-UE: Learning to Learn Nothing for Multi-Task Learning
294. ICML 2025Near-Optimal Decision Trees in a SPLIT Second
295. ICML 2025Not All Wrong is Bad: Using Adversarial Examples for Unlearning
296. ICML 2025On Explaining Equivariant Graph Networks via Improved Relevance Propagation
297. ICML 2025On the Impact of Performative Risk Minimization for Binary Random Variables
298. ICML 2025On the Vulnerability of Applying Retrieval-Augmented Generation within Knowledge-Intensive Application Domains
299. ICML 2025Optimizing Adaptive Attacks against Watermarks for Language Models
300. ICML 2025OR-Bench: An Over-Refusal Benchmark for Large Language Models

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。