Social aspects · safety papers, 2023–2026

Key findings

• Social aspects · safety went from 4 papers in 2023 to 370 in 2026, roughly 93x in three years.
• ICML publishes the most Social aspects · safety work: 364 papers from 2023 to 2026, 53% of the four-venue total.
• The fastest-growing overlap is with Alignment and Safety: papers tagged with both rose from 0 in 2023 to 137 in 2026.
• Among the 5 topics under Alignment and Responsible AI, it ranks #3 by 2026 paper count; the fastest grower in the group, Social aspects · privacy, is up 6.6x versus 93x here.

Representative papers

Spread across years and venues so every edition shows up

• 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

All Social aspects · safety papers

131 papers, page 2 of 3

51. NeurIPS 2025Learning Gradient Boosted Decision Trees with Algorithmic Recourse
52. NeurIPS 2025Learning to Steer: Input-dependent Steering for Multimodal LLMs
53. NeurIPS 2025Lifelong Safety Alignment for Language Models
54. NeurIPS 2025LLM Safety Alignment is Divergence Estimation in Disguise
55. NeurIPS 2025LLM Unlearning via Neural Activation Redirection
56. NeurIPS 2025MF-LLM: Simulating Population Decision Dynamics via a Mean-Field Large Language Model Framework
57. NeurIPS 2025MIBP-Cert: Certified Training against Data Perturbations with Mixed-Integer Bilinear Programs
58. NeurIPS 2025Minimizing False-Positive Attributions in Explanations of Non-Linear Models
59. NeurIPS 2025Monoculture or Multiplicity: Which Is It?
60. NeurIPS 2025Noise Injection Reveals Hidden Capabilities of Sandbagging Language Models
61. NeurIPS 2025One Head to Rule Them All: Amplifying LVLM Safety through a Single Critical Attention Head
62. NeurIPS 2025OOD Detection with Relative Angles
63. NeurIPS 2025OrdShap: Feature Position Importance for Sequential Black-Box Models
64. NeurIPS 2025OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents
65. NeurIPS 2025Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning
66. NeurIPS 2025Panacea: Mitigating Harmful Fine-tuning for Large Language Models via Post-fine-tuning Perturbation
67. NeurIPS 2025Performative Validity of Recourse Explanations
68. NeurIPS 2025PolyGuard: Massive Multi-Domain Safety Policy-Grounded Guardrail Dataset
69. NeurIPS 2025Predicting the Performance of Black-box Language Models with Follow-up Queries
70. NeurIPS 2025Preference Learning with Lie Detectors can Induce Honesty or Evasion
71. NeurIPS 2025Probabilistic Stability Guarantees for Feature Attributions
72. NeurIPS 2025Provable Gradient Editing of Deep Neural Networks
73. NeurIPS 2025Provably Efficient RL under Episode-Wise Safety in Constrained MDPs with Linear Function Approximation
74. NeurIPS 2025Rectifying Shortcut Behaviors in Preference-based Reward Learning
75. NeurIPS 2025Redefining Experts: Interpretable Decomposition of Language Models for Toxicity Mitigation
76. NeurIPS 2025Refusal Direction is Universal Across Safety-Aligned Languages
77. NeurIPS 2025Regional Explanations: Bridging Local and Global Variable Importance
78. NeurIPS 2025Regression Trees Know Calculus
79. NeurIPS 2025Reliably detecting model failures in deployment without labels
80. NeurIPS 2025Representational Difference Explanations
81. NeurIPS 2025Residual Stream Analysis of Overfitting And Structural Disruptions
82. NeurIPS 2025RiOSWorld: Benchmarking the Risk of Multimodal Computer-Use Agents
83. NeurIPS 2025Risk Management for Mitigating Benchmark Failure Modes: BenchRisk
84. NeurIPS 2025Robust Explanations of Graph Neural Networks via Graph Curvatures
85. NeurIPS 2025Safe RLHF-V: Safe Reinforcement Learning from Multi-modal Human Feedback
86. NeurIPS 2025SAFEPATH: Preventing Harmful Reasoning in Chain-of-Thought via Early Alignment
87. NeurIPS 2025Safety Depth in Large Language Models: A Markov Chain Perspective
88. NeurIPS 2025Safety Pretraining: Toward the Next Generation of Safe AI
89. NeurIPS 2025SafeVid: Toward Safety Aligned Video Large Multimodal Models
90. NeurIPS 2025SAFEx: Analyzing Vulnerabilities of MoE-Based LLMs via Stable Safety-critical Expert Identification
91. NeurIPS 2025SAGE-Eval: Evaluating LLMs for Systematic Generalizations of Safety Facts
92. NeurIPS 2025Sample-efficient Learning of Concepts with Theoretical Guarantees: from Data to Concepts without Interventions
93. NeurIPS 2025Scalable Fingerprinting of Large Language Models
94. NeurIPS 2025Scalable Valuation of Human Feedback through Provably Robust Model Alignment
95. NeurIPS 2025Scaling Laws For Scalable Oversight
96. NeurIPS 2025Semantic Representation Attack against Aligned Large Language Models
97. NeurIPS 2025SHAP Meets Tensor Networks: Provably Tractable Explanations with Parallelism
98. NeurIPS 2025SHAP values via sparse Fourier representation
99. NeurIPS 2025SHAP zero Explains Biological Sequence Models with Near-zero Marginal Cost for Future Queries
100. NeurIPS 2025Shape it Up! Restoring LLM Safety during Finetuning

FAQ

How many Social aspects · safety papers were published at NeurIPS, ICLR, ICML and CVPR from 2023 to 2026?

688 in total: 4 in 2023, 41 in 2024, 273 in 2025, 370 in 2026.

How many Social aspects · safety papers are at NeurIPS 2026?

NeurIPS 2026 has 121 Social aspects · safety papers, versus 131 the year before.

How many Social aspects · safety papers are at ICLR 2026?

ICLR 2026 has 20 Social aspects · safety papers, versus 11 the year before.

How many Social aspects · safety papers are at ICML 2026?

ICML 2026 has 226 Social aspects · safety papers, versus 130 the year before.

How many Social aspects · safety papers are at CVPR 2026?

CVPR 2026 has 3 Social aspects · safety papers, versus 1 the year before.

Which conference publishes the most Social aspects · safety papers?

ICML, with 364 papers from 2023 to 2026.