safe-ai
Aegis-AI-Content-Safety-LlamaGuard-Defensive-1.0Aegis-AI-Content-Safety-LlamaGuard-Permissive-1.0Model-SafeTensors-Lumimaid-v0.2-70B-GGUFsomo-olmo-7b-sdf-sfttyphoon2-safety-previewMergeBench-gemma-2-2b-it_safety-GGUFpriyanshi27dixit-SAFETY_FULL_FT_VECTOR-GGUFcookinai-LlamaReflect-8B-CoT-safetensors-GGUF
Datasets
All datasets matching “safe-ai”AIC_final_safe_red_permuted_500AIC_final_safe_red_220AIC_final_safe_red_398AIC_final_safe_red_permuted_500_labeledaqshift-us-leakage-safe-air-quality
AQShift-US
AQShift-US is a leakage-resistant benchmark built from real U.S. EPA Air
Quality System measurements. It supports next-local-day forecasting and
interval calibration under forward-time and unseen-site shift for daily
maximum 8-hour ozone and observed 24-hour PM2.5.
The frozen source snapshot covers 2010–2025 and was published by EPA AirData
on 2026-06-25. It contains 7,544,381 mature benchmark examples derived from
6,234,916 canonical ozone monitor-days and 1,447,795… See the full description on the dataset page: https://huggingface.co/datasets/haidang2405/aqshift-us-leakage-safe-air-quality.daily-paper-2026-07-16-safe-autonomy-k8s-remediation
Escalate or Act? Calibrating the Safe-Autonomy Boundary for LLM Agents in Closed-Loop Kubernetes GPU Incident Remediation
TL;DR — Calibrating a separate escalate/auto-remediate threshold per Kubernetes incident type (OOM, PVC, node pressure, scheduler) recovers 52.2% MTTR reduction at a 2% catastrophic-escape safety ceiling — 10.4 pp more than a single global threshold — because incident types differ sharply in blast radius and tenant exposure.
ThakiCloud AI Research ·… See the full description on the dataset page: https://huggingface.co/datasets/thaki-AI/daily-paper-2026-07-16-safe-autonomy-k8s-remediation.