Towards Architectural-Level Safety Control: Relocating Hazardous Knowledge into Removable Experts in Mixture-of-Experts Models
AI & ML interests
None defined yet.
Recent Activity
Papers
ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility Balancing
Organization Card
AI45Research
Welcome to AI45Lab! We are a research group from Shanghai AI Lab focused on AI safety. Our mission to co-evolve the safety and capability of AI marks the birth of the AI45 brand. With "Make Safe AI" as the core technology vision, the team focuses on strengthening the foundation of endogenous safety AI technology, promoting the industry's transformation from "Make AI Safe" to "Make Safe AI".This supports the healthy and orderly development of artificial intelligence in a beneficial, safe, and fair direction.
Models
- SafeWork-R1: a pioneering open-source safety-oriented model with good safety and capability performance.
ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
-
AI45Research/ReSI-Safin-1-35B-A3B
Text Generation • 36B • Updated • 25 • 1 -
AI45Research/ReSI-Qwen3.6-27B
Image-Text-to-Text • 27B • Updated • 21 -
AI45Research/ReSI-Ministral-3-14B-Reasoning
Image-Text-to-Text • 14B • Updated • 21 -
ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
Paper • 2610.12233 • Published • 27
Towards Architectural-Level Safety Control: Relocating Hazardous Knowledge into Removable Experts in Mixture-of-Experts Models
ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
-
AI45Research/ReSI-Safin-1-35B-A3B
Text Generation • 36B • Updated • 25 • 1 -
AI45Research/ReSI-Qwen3.6-27B
Image-Text-to-Text • 27B • Updated • 21 -
AI45Research/ReSI-Ministral-3-14B-Reasoning
Image-Text-to-Text • 14B • Updated • 21 -
ReSI: Recursive Safety Improvement toward Resistant and Resilient AI
Paper • 2610.12233 • Published • 27
models 28
AI45Research/REX-Intern-S2-Bio-F3
Image-Text-to-Text • 36B • Updated
AI45Research/REX-Intern-S2-Cyber-F3
Updated
AI45Research/ReSI-DeepSeek-V4-Flash
Text Generation • 284B • Updated • 28
AI45Research/ReSI-Ministral-3-14B-Reasoning
Image-Text-to-Text • 14B • Updated • 21
AI45Research/ReSI-Qwen3.6-27B
Image-Text-to-Text • 27B • Updated • 21
AI45Research/ReSI-Safin-1-35B-A3B
Text Generation • 36B • Updated • 25 • 1
AI45Research/AgentDoG1.5-Llama-3.1-8B
Text Classification • 8B • Updated • 59
AI45Research/AgentDoG1.5-Unified-Qwen3.5-4B
Text Classification • 4B • Updated • 75 • 3
AI45Research/AgentDoG1.5-FG-Llama-3.1-8B
Text Classification • 8B • Updated • 31 • 1
AI45Research/AgentDoG1.5-FG-Qwen3.5-4B
Text Classification • 5B • Updated • 40 • 1
datasets 7
AI45Research/SATraj-OS
Viewer • Updated • 10.9k • 747 • 4
AI45Research/2026_summer_camp_teseset
Preview • Updated • 25 • 2
AI45Research/APP1-Agentic-Safety-SFT-Data
Viewer • Updated • 78.7k • 87 • 3
AI45Research/AgentDoG1.0-Training-Data
Viewer • Updated • 8k • 401
AI45Research/ATBench
Viewer • Updated • 1.5k • 2.62k • 43
AI45Research/ATBench-Codex
Viewer • Updated • 500 • 391
AI45Research/ATBench-Claw
Viewer • Updated • 500 • 342 • 3