GuardReasoner: Towards Reasoning-based LLM Safeguards
Paper • 2501.18492 • Published • 88
AI censorship
-
-
Safeguard Fine-Tuned LLMs Through Pre- and Post-Tuning Model Merging
Paper • 2412.19512 • Published • 9 -
Course-Correction: Safety Alignment Using Synthetic Preferences
Paper • 2407.16637 • Published • 26 -
Refusal in Language Models Is Mediated by a Single Direction
Paper • 2406.11717 • Published • 9 -
GuardReasoner-VL: Safeguarding VLMs via Reinforced Reasoning
Paper • 2505.11049 • Published • 61 -
Lifelong Safety Alignment for Language Models
Paper • 2505.20259 • Published • 24 -
Automating Steering for Safe Multimodal Large Language Models
Paper • 2507.13255 • Published • 4 -
The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMs
Paper • 2507.11097 • Published • 64 -
T2VSafetyBench: Evaluating the Safety of Text-to-Video Generative Models
Paper • 2407.05965 • Published -
Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report
Paper • 2507.16534 • Published • 9 -
AlignGuard-LoRA: Alignment-Preserving Fine-Tuning via Fisher-Guided Decomposition and Riemannian-Geodesic Collision Regularization
Paper • 2508.02079 • Published • 3 -
Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs
Paper • 2508.06601 • Published • 7 -
Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection
Paper • 2508.20766 • Published • 14 -
An Embarrassingly Simple Defense Against LLM Abliteration Attacks
Paper • 2505.19056 • Published • 6 -
The Rogue Scalpel: Activation Steering Compromises LLM Safety
Paper • 2509.22067 • Published • 28 -
Imperceptible Jailbreaking against Large Language Models
Paper • 2510.05025 • Published • 33 -
Too Good to be Bad: On the Failure of LLMs to Role-Play Villains
Paper • 2511.04962 • Published • 57 -
In-Context Representation Hijacking
Paper • 2512.03771 • Published • 5 -
OmniSafeBench-MM: A Unified Benchmark and Toolbox for Multimodal Jailbreak Attack-Defense Evaluation
Paper • 2512.06589 • Published • 19 -
GateBreaker: Gate-Guided Attacks on Mixture-of-Expert LLMs
Paper • 2512.21008 • Published • 4 -
Safety at One Shot: Patching Fine-Tuned LLMs with A Single Instance
Paper • 2601.01887 • Published • 1 -
A Safety Report on GPT-5.2, Gemini 3 Pro, Qwen3-VL, Doubao 1.8, Grok 4.1 Fast, Nano Banana Pro, and Seedream 4.5
Paper • 2601.10527 • Published • 26 -
The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models
Paper • 2601.10387 • Published • 15 -
Selective Steering: Norm-Preserving Control Through Discriminative Layer Selection
Paper • 2601.19375 • Published • 5 -
THINKSAFE: Self-Generated Safety Alignment for Reasoning Models
Paper • 2601.23143 • Published • 39 -
Large Language Lobotomy: Jailbreaking Mixture-of-Experts via Expert Silencing
Paper • 2602.08741 • Published • 2 -
UGI Leaderboard
📢1.7kUncensored General Intelligence Leaderboard
-
Intent Laundering: AI Safety Datasets Are Not What They Seem
Paper • 2602.16729 • Published • 1 -
SOM Directions are Better than One: Multi-Directional Refusal Suppression in Language Models
Paper • 2511.08379 • Published • 4 -
Large Language Models Generate Harmful Content Using a Distinct, Unified Mechanism
Paper • 2604.09544 • Published • 6 -
ASGuard: Activation-Scaling Guard to Mitigate Targeted Jailbreaking Attack
Paper • 2509.25843 • Published • 19