safety

When Noise Meets Long-Tail: Feature-Threshold Dual Calibration for Robust Pseudo-Labeling

arXiv:2609.33668v1 Announce Type: new Abstract: Pseudo-labeling has become a cornerstone of learning from unlabeled data in semantic segmentation. Yet its effectiveness drops sharply in real-world sce

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Safety

When Should a Human Take Back Control? Optimal Delegation under Turbulent AI Risk

DGX agent

arXiv:2609.32083v1 Announce Type: new Abstract: Deploying AI systems requires deciding when to delegate tasks and when humans should intervene to monitor and mitigate risk induced by AI operations. Th

safetyarxiv-cs-lg

29 Sept 2026

When Sparse Reward Meets Dense Distillation: Training Dynamics of On-Policy Distillation

arXiv:2609.34849v1 Announce Type: new Abstract: Reinforcement learning with verifiable rewards provides a sparse post-training signal: a single binary outcome evaluates the entire rollout, and every t

When to Commit and When to Defer: Maturing Markov Decision Processes under Refining Information and Expiring Opportunities

arXiv:2606.18820v2 Announce Type: replace Abstract: Sequential decisions often become better informed as action opportunities expire. We introduce Maturing Markov Decision Processes (MMDPs), a structu

When World Models Lie: Adaptive Safety Analysis Under Wrong Imaginations

arXiv:2609.34300v1 Announce Type: new Abstract: World models offer a powerful substrate for safety reasoning in high-dimensional robotic systems, but they are also fallible: their predictions can be b

Where Activation Sparsity and KV-Cache Sparsity Cross in LLM Decoding

arXiv:2609.33889v1 Announce Type: new Abstract: At each step, decoding one sequence with a large language model rereads the projection weights, whose traffic is fixed, and the key-value (KV) cache, wh

Where Memory Belongs: Ledger, an Object Ledger for Memory-Augmented VLAs

arXiv:2609.34554v1 Announce Type: new Abstract: Memory is essential for long-horizon, partially observed robotic manipulation: a robot must remember which object was placed in a drawer, whose cup it m

WorldGuide: Learning Success-Failure Boundaries in Latent World Models for Vision-Language-Action Policies

arXiv:2609.34206v1 Announce Type: new Abstract: Latent world models offer a promising way to improve Vision-Language-Action policies by capturing the consequences of actions. However, models trained p

WSM-Aware HRI: An IoT-Enhanced Framework for Early Detection and Norm-Guided Repair of Failures with LLM Guidance

arXiv:2609.32336v1 Announce Type: new Abstract: Human-robot interaction (HRI) failures remain a major barrier to deploying robots in real-world environments. Prior work often treats failures as isolat

X-Reset: Scaling Object-Centric Reinforcement Learning via Cross-Embodiment Resets

arXiv:2609.35715v1 Announce Type: new Abstract: Reinforcement learning (RL) in simulation can train dexterous manipulation policies without robot demonstrations, but training a single generalist polic

You always stay in control. Set boundaries and specify what your dot can do on its own, when it should ask first, and what it should never d…

You always stay in control. Set boundaries and specify what your dot can do on its own, when it should ask first, and what it should never do. Safety is built in. You choose which apps to connect, and

safetyopenai--x

Zero-Shot Reactive Obstacle Avoidance for Generative Robot Policies

arXiv:2609.35231v1 Announce Type: new Abstract: We propose NUDGE (Nudge Update via Differentiable GEometry), a training-free obstacle-avoidance procedure that can be incorporated in any robot policy b

ZeroBot: Learning from Scratch in Minutes with Generative Real2Sim

arXiv:2609.34010v1 Announce Type: cross Abstract: We present ZeroBot, a real2sim framework for learning a robot manipulation task from scratch in minutes under challenging conditions: zero human demon

ZeroCode: On-demand Error-Correcting Code Construction from the Zero Matrix via Reinforcement Learning

arXiv:2609.34265v1 Announce Type: new Abstract: Error-correcting codes (ECCs) are essential across diverse applications, from wireless communications and storage to quantum computing, yet each applica

ZonoGPT: Towards An Abstract Domain for Verifying Large GPT Models

arXiv:2609.34457v1 Announce Type: new Abstract: Transformer-based models are widely used for reasoning, coding, and multimodal agentic tasks. To provide formal assurance of desirable behaviors, such a

A Safety-Bounded SDC-to-MCP Gateway for Medical AI Agents

arXiv:2609.31358v1 Announce Type: cross Abstract: The Model Context Protocol (MCP) provides a common interface through which AI applications discover and use external resources and tools. It allows la

Accounting for Bias Enables Sustainable LLM Evaluation

arXiv:2609.31184v1 Announce Type: new Abstract: LLM-as-a-judge has become the de facto standard for scalable, subjective evaluation, yet current leaderboards compensate for systematic measurement bias

Actively Resolving Contextual Uncertainty for Underspecified Tasks in Natural Language

arXiv:2609.30428v1 Announce Type: cross Abstract: Foundation models provide robots with the ability to interpret natural language and reason about environmental context, yet most language-conditioned

Adaptive Multi-Value Control in LLMs via Causal Activation Steering

arXiv:2609.30405v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed in settings where responses must reflect multiple, potentially interacting social norms and human

Adaptive Random Matrices in Gaussian Bandits: Spectral Universality and Selection-Induced Outliers

arXiv:2609.30321v1 Announce Type: cross Abstract: Adaptive arm selection changes the distribution of the observations collected by a bandit algorithm, but it need not change their limiting empirical s

Add Runtime Controls to AI Agents with NVIDIA OpenShell

NVIDIA OpenShell 0.1.0 is an open‑source runtime that enforces which systems and data an AI agent may access without requiring changes to the agent itself. Its components—Gateway, Supervisor, Sandbox,

safetynvidia-developer

Aerial Manipulation in the Wild with Onboard Perception, Policy Learning, and Whole-Body Control

arXiv:2609.30521v1 Announce Type: new Abstract: Aerial manipulation in outdoor environments remains challenging due to the simultaneous requirements of reliable state estimation, stable aerial motion,

Aligning One-Step Generative Models with Reward-Weighted Transport Distillation

arXiv:2609.30840v1 Announce Type: cross Abstract: One-step generators enable high-quality visual generation with a single network evaluation, but their post-training is difficult: general implicit gen

All In Good Time: Causality-Aware Framework for LLM-Based Simultaneous Speech-to-Speech Translation

arXiv:2609.30416v1 Announce Type: new Abstract: Large Language Models (LLMs) have shown strong performance in low-resource offline translation; however, extending them to simultaneous speech-to-speech

safetyarxiv-cs-cl

always the CEOs, very rarely independent scientists. and that, my friends, is how regulatory capture works

always the CEOs, very rarely independent scientists. and that, my friends, is how regulatory capture works So far top attendees tomorrow I've confirmed for the Trump/Johnson/AI meeting are Dario Amode

safetygary-marcus--x

Amplify What You Gaze At: Target Saliency Boosting in Text-to-Image Generation

arXiv:2609.30733v1 Announce Type: new Abstract: Text-to-image generation has advanced in controlling what, where, and how objects appear, yet how visual attention is distributed among objects remains

An End-to-End Pipeline for Causal ML with Continuous Treatments: An Application to Financial Decision Making

arXiv:2609.30396v1 Announce Type: cross Abstract: This paper presents an end-to-end causal machine learning (ML) pipeline designed for real-world applications with continuous treatments. The proposed

An OpenAI agent security executive discusses the Hugging Face incident, OpenAI's response, sandboxing improvements, alignment, 'reasonable paranoia', and more (Joe/@joedaroo)

Joe / @joedaroo: An OpenAI agent security executive discusses the Hugging Face incident, OpenAI's response, sandboxing improvements, alignment, “reasonable paranoia”, and more — Took a minute to write

safetytechmeme

Anchor to Expand: Semantic Anchoring for Personalized Text-to-Image Diffusion Models

arXiv:2511.22245v2 Announce Type: replace Abstract: Personalizing text-to-image diffusion models extends pretrained models to represent novel user-specific concepts from only a few reference images. H

Attention-Based Adaptive Policies for Simultaneous Speech-to-Text Translation

arXiv:2609.30839v1 Announce Type: new Abstract: Simultaneous speech-to-text translation (Simul-S2TT) consists of generating partial translations while the incoming audio frames are processed by the sy

Audio LLMs Know When They Can't Hear You

arXiv:2609.30625v1 Announce Type: new Abstract: Audio large language models allow users to interact with the model through speech. When an input recording is too degraded, the model may misinterpret t

AxonSynth: Domain-Randomized Synthetic Data for Zero-Shot 3D Axon Segmentation in Light-Sheet Microscopy

arXiv:2609.31431v1 Announce Type: new Abstract: Accurate segmentation of axons in 3D microscopy data is important for analyzing white-matter organization, but dense ground truth labels are expensive t

BREAKING: Florida becomes the first state to do what I have been recommending for weeks!

BREAKING: Florida becomes the first state to do what I have been recommending for weeks! NEWS: Florida is calling for a pause in the development of ChatGPT, citing my @axios scoop on top AI labs inves

safetygary-marcus--x

Bridging Body and Brain: Gene-Driven Morphology--Control Co-Design

arXiv:2609.31329v1 Announce Type: new Abstract: Morphology--control co-design jointly optimizes an agent's body structure and control policy as an integrated embodied system. However, existing methods

Bridging LLM Agents and Data Spaces: An Architectural Mediation Approach using the Model Context Protocol

arXiv:2609.30341v1 Announce Type: new Abstract: Data Spaces enable sovereign and governed data sharing across organizational boundaries, but their integration with AI agents remains challenging due to

Building Trust Into Agentic AI: Why Vultr Is Evaluating the NVIDIA Agent Safety Platform

Vultr is evaluating NVIDIA’s Agent Safety Platform as enterprise agents move from experimentation to production and receive real autonomy over internal systems. The platform addresses trust by securin

safetyvultr

Bundled Contact Gradients: Stabilizing Differentiable Simulation for Deployable Dynamic Tasks

arXiv:2609.30951v1 Announce Type: new Abstract: Differentiable simulation provides analytic gradients of robot dynamics, enabling fast and sample-efficient first-order policy optimization. However, ob

Can a Robot Read Braille? - Learning to Adapt Contact via Imitation Learning for Tactile Braille Recognition

arXiv:2609.30676v1 Announce Type: new Abstract: For people who are blind, touch provides an essen-tial channel for accessing written information through Braille. Bringing a similar capability to robot

CG-Probes: Recovering Guardrail Directions from Patient Query Embeddings

arXiv:2609.31062v1 Announce Type: new Abstract: Patient-facing AI assistants promise valuable support to patients, but incoming queries can pose medical risks. To create guardrails, we work with oncol

safetyarxiv-cs-cl

Common-Mode Collapse and Recovery in Direct Feedback Alignment

arXiv:2609.31589v1 Announce Type: new Abstract: Direct feedback alignment (DFA) trains hidden layers through fixed random projections of output error. With tanh hidden units and independent sigmoid ou

Conformal Prediction under Exponential-Tilt Joint Shift

arXiv:2609.30886v1 Announce Type: cross Abstract: Conformal prediction can lose coverage when the data distribution changes after deployment. We study adaptation using labeled source data and unlabele

Counterfactual Online Conformal Prediction Under Adaptive Logging

arXiv:2609.30811v1 Announce Type: new Abstract: Online conformal prediction can fail when predictions shape actions and actions determine which outcomes enter calibration. Standard adaptive methods ma

Demo: Generative AI helps Radiotherapy Planning with User Preference

arXiv:2512.08996v2 Announce Type: replace-cross Abstract: Radiotherapy planning is a highly complex process that often varies significantly across institutions and individual planners. Most existing d

Deterministic Regime Switching and Feasibility Inversion in Dynamic Tensor Rematerialization

arXiv:2609.31250v1 Announce Type: new Abstract: We report fine-grained, deterministic instability in Dynamic Tensor Rematerialization (DTR), an online eviction policy for memory-constrained DNN traini

DIAL: Position-Debiased LLM Judges with Adaptive Human Preference Calibration

arXiv:2609.31215v1 Announce Type: new Abstract: Large language models (LLMs) as a judge enable scalable evaluation, but their judgments can be sensitive to response order and, even after removing such

Do we need to answer that question? Salience and Answerability of Potential Questions in Naturalistic Dialogue

arXiv:2609.31130v1 Announce Type: new Abstract: We empirically investigate Question Under Discussion based modelling in naturalistic dialogue by studying whether the salience of generated potential qu

safetyarxiv-cs-cl

Dynamic Welfare-Maximizing Pooled Testing

arXiv:2601.22419v2 Announce Type: replace-cross Abstract: Pooled testing uses one test to certify several agents as healthy when the pooled result is negative. We study a budget-constrained welfare pr

Electric Vehicle Charging Station Location Selection using Geospatial Artificial Intelligence (GeoAI)

arXiv:2609.30417v1 Announce Type: new Abstract: As electric vehicle (EV) adoption increases, ensuring efficient and well-distributed charging infrastructure has become a critical challenge. While many