safety

Precision autotuning for linear solvers via contextual bandit-based RL

arXiv:2601.00728v5 Announce Type: replace Abstract: We propose a reinforcement learning (RL) framework for xy{responsive} precision tuning for linear solvers, which can be extended to general algorith

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

QUATRO: Query-Adaptive Trust Region Policy Optimization for LLM Fine-tuning

DGX agent

arXiv:2602.04620v3 Announce Type: replace Abstract: GRPO-style reinforcement learning (RL)-based LLM fine-tuning algorithms have recently gained popularity. Relying on heuristic trust-region approxima

safetyarxiv-cs-lg

18 Sept 2026

Recency Forcing: Bridging the Long-Horizon Gap in Autoregressive Video Generation

arXiv:2609.19729v1 Announce Type: new Abstract: Autoregressive (AR) video generation degrades over long horizons due to an overlooked train-inference discrepancy we term KV eviction mismatch: models t

Refuse, Decompose, Refresh: A Claim-Safe Protocol for Closed-Loop AI Evaluation

arXiv:2609.20538v1 Announce Type: new Abstract: An AI evaluation can be perfectly reproducible and still support the wrong claim. This risk is acute in closed-loop systems: policy determines visited s

Regularized Emphatic Temporal-Difference Learning: Stability under Constant Stepsizes

arXiv:2609.19170v1 Announce Type: new Abstract: Emphatic temporal-difference learning (ETD) stabilizes the expected off-policy TD update and changes its projection geometry, but neither property deter

Reproducibility is not construct validity: LLM measurement of institutionally situated communication

arXiv:2609.19866v1 Announce Type: new Abstract: High annotation reproducibility does not necessarily imply that an LLM-inferred measure captures the construct it is intended to measure. We test this d

ReShoot: Generative Visual Domain Randomization of Recorded Robot Demonstrations for Visuomotor Policy Learning

arXiv:2609.19661v1 Announce Type: new Abstract: Imitation-learned robot policies are frequently overfit to the visual conditions present in their training demonstrations. Consequently, variations in o

RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning

arXiv:2609.20784v1 Announce Type: cross Abstract: Multi-turn agents trained with reinforcement learning (RL) receive a single scalar reward per trajectory, which motivates self on-policy distillation

Risk-Set Transported Synthetic Control with Difference-in-Differences Adjustment under Staggered Treatment Adoption

arXiv:2609.20264v1 Announce Type: cross Abstract: In staggered treatment-adoption designs, later-treated units are valid controls for an earlier-treated cohort only until their own treatment begins, s

Robotic Video World Models: A Survey of Applications, Research Challenges, Future Directions

arXiv:2601.07823v2 Announce Type: replace-cross Abstract: Video world models have emerged as promising candidates for high-fidelity world models, offering the potential to synthesize high-quality vide

Robust Workflow Generation via Adversarial Learning for Audio Deepfake Detection

arXiv:2609.20063v1 Announce Type: cross Abstract: The rapid advancement of speech synthesis and voice conversion technologies has made audio deepfakes increasingly realistic, posing serious security r

RotateIt! Fast and Reliable Single-Arm Garment Unfolding via Online-Adaptive Dynamic Rotation

arXiv:2609.19817v1 Announce Type: new Abstract: Robotic garment unfolding is essential for downstream tasks, yet quasi-static methods require repeated actions, while existing dynamic approaches predom

RTK-Vision PPO for Autonomous Micro UAV Recovery on an Airborne Carrier

arXiv:2609.20629v1 Announce Type: new Abstract: Autonomous recovery of a micro unmanned aerial vehicle (UAV) onto a moving airborne carrier enables reusable deploy-mission-recover operation, but coupl

Runtime Safety Filtering for Two-Terminal Hazards in Robotic Battery Recycling

arXiv:2609.19665v1 Announce Type: new Abstract: Runtime safety filters for learned manipulation policies typically define unsafe states as unions of object-wise keep-out regions. This representation c

Safety-Critical Scenanrio Emerges from Initial Scene

arXiv:2609.20103v1 Announce Type: new Abstract: Safety-critical driving scenario generation has largely focused on manipulating the behavior of surrounding agents while starting from an initial scene

Schema-Anchored Latent Reasoning for Semantic Parsing-Based Knowledge Base Question Answering

arXiv:2609.20398v1 Announce Type: new Abstract: Semantic parsing (SP)-based knowledge base question answering aims to answer natural language questions by generating executable logical forms (LFs) ove

Score Centering Stabilizes Off-policy Reinforcement Learning

arXiv:2609.20807v1 Announce Type: new Abstract: Reinforcement learning (RL) of large language models is notoriously sensitive to small differences between training and inference engines, often referre

Seeing Abnormal from Normal: Glomerular Abnormality in Representations of Normal Renal Morphology

arXiv:2609.19444v1 Announce Type: new Abstract: Fine-grained evaluation of glomerular pathology must distinguish normal glomeruli from abnormalities such as global and segmental glomerulosclerosis, ob

SemSafe-3DGS: Semantic Risk-Aware Active Navigation in Uncertain 3D Gaussian Splatting Maps

arXiv:2609.19330v1 Announce Type: new Abstract: Autonomous robots operating in partially observed environments must navigate safely while acquiring observations that improve future planning. Existing

SGM: A Statistical Godel Machine for Risk-Controlled Recursive Self-Modification

arXiv:2510.10232v2 Announce Type: replace-cross Abstract: Recursive self-modification is increasingly central in AutoML, neural architecture search, and adaptive optimization, yet no existing framewor

SkipVLA: Skipping VLA Steps with Classical Planning for Fast Robot Manipulation

arXiv:2609.20648v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models are a class of generalist robot policies that map camera images and language instructions directly to robot actions.

Sources: Anthropic expects to generate 100B+ in annualized revenue this year, up from 65B as of July, as it moves ahead with its IPO amid the AI safety debate (New York Times)

New York Times: Sources: Anthropic expects to generate 100B+ in annualized revenue this year, up from 65B as of July, as it moves ahead with its IPO amid the AI safety debate — The company is expected

safetytechmeme

Stable Policy Learning

arXiv:2609.19418v1 Announce Type: cross Abstract: In evidence-based policymaking, typically one experimental sample is observed, then a learned policy recommendation is implemented at scale. Policies

StageGuard: Learning Stage Transitions for Long-Horizon Robot Tasks via Agentic Distillation

arXiv:2609.20791v1 Announce Type: new Abstract: Hierarchical planning frameworks combine skills from multiple robot control policies for long-horizon task execution, where determining when to terminat

State-Change Learning for Prediction of Future Events in Endoscopic Videos

arXiv:2510.12904v2 Announce Type: replace Abstract: Surgical future prediction, driven by real-time AI analysis of surgical video, is critical for operating room safety and efficiency. It provides act

Steering Equilibrium Selection in Regularized Self-Play via the Reference Policy

arXiv:2609.19820v1 Announce Type: new Abstract: Regularized self-play -- the family behind DeepNash's Stratego play -- drives a two-player zero-sum policy to a Nash equilibrium by best-responding to a

STR-Agent: An LLM-Driven Agent for QoS-Aware Routing in LEO Satellite Networks

arXiv:2609.20347v1 Announce Type: cross Abstract: LEO satellite networks feature dynamic topologies, time-varying links, and diverse service requirements, which make conventional routing schemes diffi

Summarization Bias: The Directional Collapse of Objective Projection into Told-Mode Labels in Large Language Models --- A Conceptual Framework and Registered Test Protocol

arXiv:2609.20712v1 Announce Type: new Abstract: This paper introduces and operationalizes summarization bias: a proposed systematic tendency of large language models (LLMs) to represent narrative mean

TacSushi: Tactile-Grounded World-Action Modeling for Dexterous Sushi Manipulation

arXiv:2609.19613v1 Announce Type: cross Abstract: Dexterous food manipulation requires control under deformation, occlusion, and uncertain contact. We present TacSushi, a tactile-grounded, Cosmos3-bas

Tele-Traversability: Rethinking Traversability for Teleoperated Ground Robots in Terrain Navigation

arXiv:2609.19577v1 Announce Type: new Abstract: Teleoperation, a human-in-the-loop control scheme, allows a human operator to remotely command and guide a mobile robot to navigate in off-road environm

The AR Fairness Metamodel: A Structured Framework for Fairness Measures

arXiv:2609.19234v1 Announce Type: cross Abstract: This paper presents the AR fairness metamodel, a framework designed to represent, analyze, and compare different fairness scenarios. The metamodel con

The Bias of Nonlinear Two-Time-scale Stochastic Approximation under Constant Step-Sizes

arXiv:2609.20409v1 Announce Type: new Abstract: Two-timescale stochastic approximation (TTSA) is a fundamental tool for analyzing coupled iterative algorithms in reinforcement learning, optimization,

The Role of Fine-grained Harm Signals in LLM Safety

arXiv:2609.19366v1 Announce Type: new Abstract: Prior work has shown that internal harmfulness representations in large language models vary across risk categories, while sharing a common general harm

These are precisely the scenarios that I spend a significant time actually being concerned with. The adoption of unreliable, brittle, and in…

These are precisely the scenarios that I spend a significant time actually being concerned with. The adoption of unreliable, brittle, and inaccurate LLMs within safety-critical fields (e.g., defense a

safetygary-marcus--x

Think Thrice Before Reranking: Multi-perspective Evidence and Reasoning Integration for Text Reranking

arXiv:2609.20131v1 Announce Type: cross Abstract: Reasoning-based reranking with Large Language Models (LLMs) has shown promising improvements in text ranking. However, current methods predominantly r

this this this

this this this The focus on AI causing extinction is going to get people killed. AI lab employees talk about how hard security is, not to point to some new solution, but to say security won’t work. On

safetygary-marcus--x

This was a super fun podcast, went far further afield than most of them do. Give it a listen!

This was a super fun podcast, went far further afield than most of them do. Give it a listen! Episode 432 w/ Connor Leahy (@NPCollapse) is available now. Connor is an AI researcher & co-founder of Ele

safetyconnor-leahy--x

Time-Efficient Iterative Learning Planning for Safety-Critical Dynamic Obstacle Avoidance

arXiv:2609.20435v1 Announce Type: new Abstract: Autonomous mobile robots require timeefficient planning and safety-critical dynamic obstacle avoidance under constrained onboard computation. While Iter

Towards High-DoF Dexterous Manipulation through VLA Post-Training

arXiv:2609.19666v1 Announce Type: new Abstract: Imitation-learned vision--language--action (VLA) foundation models acquire broad manipulation capabilities by scaling robot data across tasks and embodi

TraceFlow: Guiding Frozen Flow-Matching Robot Policies with Success and Failure Traces

arXiv:2609.20646v1 Announce Type: new Abstract: A vision-language-action (VLA) policy with a flow-matching action expert generates each action chunk (a short command sequence) by integrating a learned

ULOHA: An Underwater Bimanual Robot System for Robot Learning

arXiv:2609.19200v1 Announce Type: new Abstract: Underwater visuomotor policy learning has focused primarily on single manipulators, while bimanual imitation learning has been studied largely in air. W

UniExo: Unified Multi-Skill Policies for Musculoskeletal Locomotion and Co-Adaptive Exoskeleton Control

arXiv:2609.19690v1 Announce Type: cross Abstract: Daily locomotion encompasses diverse activities and frequent transitions between them, yet most exoskeleton controllers are designed for a single acti

VGGT-GS SLAM: Uncalibrated Monocular Gaussian Splatting SLAM with Feed-Forward Priors

arXiv:2609.19628v1 Announce Type: new Abstract: We present VGGT-GS SLAM, a monocular 3D Gaussian Splatting SLAM system designed for uncalibrated videos. Starting from feed-forward VGGT pose and depth

Video DeltaNet: A Video-Native Hybrid Attention for Livestream Video Generation

arXiv:2609.20744v1 Announce Type: new Abstract: Video diffusion models repeatedly process long spatiotemporal token sequences during denoising, making attention a major computational bottleneck. Linea

ViLoMan: Learning Visual-Proprioceptive Whole-Body Loco-Manipulation Skills for Humanoid Robots

arXiv:2609.19340v1 Announce Type: new Abstract: Humanoid loco-manipulation requires adaptive whole-body coordination to seamlessly integrate locomotion and physical interaction. Despite recent advance

Visual Sim-to-Real Learning for Robotic Insertion under Geometric Variations: Application to Rebar Installation

arXiv:2609.20477v1 Announce Type: new Abstract: Rebar insertion is among the most repetitive and physically demanding tasks on construction sites, and a contact-rich problem at 1.4 mm clearance. The p

VLN on the Fly: An Onboard Vision-Language Navigation Stack for Aerial Robots

arXiv:2609.20191v1 Announce Type: cross Abstract: Running vision-language navigation fully onboard an aerial robot is hard, since grounding, planning, and control must share limited compute and a sing

What Hollywood thinks about existential AI warnings

As the tech sector sounds alarms about AI's potential to destroy humanity, entertainment labor groups are urging the public to stay focused on what's already happening. The Verge reached out to Disney

safetythe-verge-ai