safety

Multi-Dimensional Model Integrity and Responsibility Assessment Index and Scoring Framework

arXiv:2605.14550v1 Announce Type: new Abstract: Artificial intelligence in high-stakes tabular domains cannot be evaluated by predictive performance alone, yet current practice still assesses explaina

Safety

Multi-scale Coarse-to-fine Modeling for Test-time Human Motion Control

DGX agent

arXiv:2605.14935v1 Announce Type: new Abstract: We present MSCoT, a multi-scale, coarse-to-fine model for test-time human motion synthesis and control. Unlike recent approaches that rely on multiple i

safetyarxiv-cs-cv

15 May 2026

Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning

arXiv:2512.07461v3 Announce Type: replace Abstract: We introduce Native Parallel Reasoner (NPR), a teacher-free framework that enables Large Language Models (LLMs) to self-evolve genuine parallel reas

NEST: Nested Event Stream Transformer for Sequences of Multisets

arXiv:2602.00520v3 Announce Type: replace Abstract: Event stream data often exhibit hierarchical structure in which multiple events co-occur, resulting in a sequence of multisets (i.e., bags of events

Not All Timesteps Matter Equally: Selective Alignment Knowledge Distillation for Spiking Neural Networks

arXiv:2605.14252v1 Announce Type: cross Abstract: Spiking neural networks (SNNs), which are brain-inspired and spike-driven, achieve high energy efficiency. However, a performance gap between SNNs and

Novel Dynamic Batch-Sensitive Adam Optimiser for Vehicular Accident Injury Severity Prediction

arXiv:2605.15083v1 Announce Type: cross Abstract: The choice of optimiser is important in deep learning, as it strongly influences model efficiency and speed of convergence. However, many commonly use

On Strong Equivalence Notions in Logic Programming and Abstract Argumentation

arXiv:2605.14721v1 Announce Type: new Abstract: Strong equivalence between knowledge bases ensures the possibility of replacing one with the other without affecting reasoning outcomes, in any given co

On the Burden of Achieving Fairness in Conformal Prediction

arXiv:2605.14260v1 Announce Type: cross Abstract: Conformal prediction is often calibrated with a single pooled threshold, but this can hide cross-group heterogeneity in score distributions and distor

On the Unreasonable Effectiveness of Last-layer Retraining

arXiv:2512.01766v2 Announce Type: replace Abstract: Last-layer retraining (LLR) methods -- wherein the last layer of a neural network is reinitialized and retrained on a held-out set following ERM tra

One Step to the Side: Why Defenses Against Malicious Finetuning Fail Under Adaptive Adversaries

arXiv:2605.14605v1 Announce Type: cross Abstract: Model providers increasingly release open weights or allow users to fine-tune foundation models through APIs. Although these models are safety-aligned

OneSearch-V2: The Latent Reasoning Enhanced Self-distillation Generative Search Framework

arXiv:2603.24422v2 Announce Type: replace-cross Abstract: Generative Retrieval (GR) has emerged as a promising paradigm for modern search systems. Compared to multi-stage cascaded architecture, it off

OpenAI's disavowal of a liability shield in Illinois SB 3444 bill and endorsement of a stronger SB 315 suggest it is open to meaningful AI safety legislation (Transformer)

Transformer: OpenAI's disavowal of a liability shield in Illinois SB 3444 bill and endorsement of a stronger SB 315 suggest it is open to meaningful AI safety legislation — Transformer Weekly: US-Chin

safetytechmeme

Performance-Driven Policy Optimization for Speculative Decoding with Adaptive Windowing

arXiv:2605.14978v1 Announce Type: new Abstract: Speculative decoding accelerates LLM inference by having a lightweight draft model propose speculative windows of candidate tokens for parallel verifica

Persian MusicGen: A Large-Scale Dataset and Culturally-Aware Generative Model for Persian Music

arXiv:2605.14765v1 Announce Type: cross Abstract: Persian music, with its unique tonalities, modal systems (Dastgah), and rhythmic structures, presents significant challenges for music generation mode

Phylogenetic Tree Inference with Tropical Axial Attention

arXiv:2605.13894v1 Announce Type: cross Abstract: In this work, we introduce a Tropical Axial Attention neural reasoning architecture that replaces vanilla softmax dot-product attention with max-plus

Policy Optimization in Hybrid Discrete-Continuous Action Spaces via Mixed Gradients

arXiv:2605.14297v1 Announce Type: cross Abstract: We study reinforcement learning in hybrid discrete-continuous action spaces, such as settings where the discrete component selects a regime (or index)

Position: Behavioural Assurance Cannot Verify the Safety Claims Governance Now Demands

arXiv:2605.15164v1 Announce Type: cross Abstract: This position paper argues that behavioural assurance, even when carefully designed, is being asked to carry safety claims it cannot verify. AI govern

Precise Verification of Transformers through ReLU-Catalyzed Abstraction Refinement

arXiv:2605.14294v1 Announce Type: new Abstract: Formal verification of transformers has become increasingly important due to their widespread deployment in safety-critical applications. Compared to cl

Pro-DG: Procedural Diffusion Guidance for Architectural Facade Generation

arXiv:2504.01571v2 Announce Type: replace-cross Abstract: We use hierarchical procedural rules for the generation of control maps within the stable diffusion framework to produce photo-realistic archi

Probabilistic Verification of Recurrent Neural Networks for Single and Multi-Agent Reinforcement Learning

arXiv:2605.14758v1 Announce Type: new Abstract: History-dependent policies induced by recurrent neural networks (RNNs) rely on latent hidden state dynamics, making verification in partially observable

Progent: Securing AI Agents with Privilege Control

arXiv:2504.11703v3 Announce Type: replace-cross Abstract: AI agents interact with external environments through tool calls, exposing them to attacks like indirect prompt injection that can trigger una

Prompting Policies for Multi-step Reasoning and Tool-Use in Black-box LLMs with Iterative Distillation of Experience

arXiv:2605.14443v1 Announce Type: new Abstract: The shift toward interacting with frozen, 'black-box' Large Language Models (LLMs) has transformed prompt engineering from a heuristic exercise into a c

Proximal Action Replacement for Behavior Cloning Actor-Critic in Offline Reinforcement Learning

arXiv:2602.07441v2 Announce Type: replace-cross Abstract: Offline reinforcement learning (RL), which optimizes policies using a previously collected static dataset, is an important branch of RL. A pop

Proxy Compression for Language Modeling

arXiv:2602.04289v2 Announce Type: replace Abstract: Modern language models are trained almost exclusively on token sequences produced by a fixed tokenizer, an external lossless compressor often over U

Quantifying and Mitigating Premature Closure in Frontier LLMs

arXiv:2605.15000v1 Announce Type: cross Abstract: Premature closure, or committing to a conclusion before sufficient information is available, is a recognized contributor to diagnostic error but remai

Quantitative Video World Model Evaluation for Geometric-Consistency

arXiv:2605.15185v1 Announce Type: cross Abstract: Generative video models are increasingly studied as implicit world models, yet evaluating whether they produce physically plausible 3D structure and m

R2PS: Worst-Case Robust Real-Time Pursuit Strategies under Partial Observability

arXiv:2511.17367v2 Announce Type: replace Abstract: Computing worst-case robust strategies in pursuit-evasion games (PEGs) is time-consuming, especially when real-world factors like partial observabil

R2R2: Robust Representation for Intensive Experience Reuse via Redundancy Reduction in Self-Predictive Learning

arXiv:2605.14026v1 Announce Type: cross Abstract: For reinforcement learning in data-scarce domains like real-world robotics, intensive data reuse enhances efficiency but induces overfitting. While pr

RAPO++: Cross-Stage Prompt Optimization for Text-to-Video Generation via Data Alignment and Test-Time Scaling

arXiv:2510.20206v2 Announce Type: replace Abstract: Prompt design plays a crucial role in text-to-video (T2V) generation, yet user-provided prompts are often short, unstructured, and misaligned with t

RAVEN: Real-time Autoregressive Video Extrapolation with Consistency-model GRPO

arXiv:2605.15190v1 Announce Type: new Abstract: Causal autoregressive video diffusion models support real-time streaming generation by extrapolating future chunks from previously generated content. Di

Reasoning Model Is Superior LLM-Judge, Yet Suffers from Biases

arXiv:2601.03630v2 Announce Type: replace Abstract: This paper presents the first systematic comparison investigating whether Large Reasoning Models (LRMs) are superior judges to non-reasoning LLMs. O

Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages

arXiv:2603.12554v2 Announce Type: replace-cross Abstract: Reinforcement learning (RL) has been effective for post-training autoregressive (AR) language models, but extending these methods to diffusion

Reinforcement Learning with Semantic Rewards Enables Low-Resource Language Expansion without Alignment Tax

arXiv:2605.14366v1 Announce Type: new Abstract: Extending large language models (LLMs) to low-resource languages often incurs an 'alignment tax': improvements in the target language come at the cost o

Resolving Action Bottleneck: Agentic Reinforcement Learning Informed by Token-Level Energy

arXiv:2605.14558v1 Announce Type: cross Abstract: Agentic reinforcement learning trains large language models using multi-turn trajectories that interleave long reasoning traces with short environment

Rethinking Output Alignment For 1-bit Post-Training Quantization of Large Language Models

arXiv:2512.21651v2 Announce Type: replace Abstract: Large Language Models (LLMs) deliver strong performance across a wide range of NLP tasks, but their massive sizes hinder deployment on resource-cons

REVISOR: Beyond Textual Reflection, Towards Multimodal Introspective Reasoning in Long-Form Video Understanding

arXiv:2511.13026v3 Announce Type: replace Abstract: Self-reflection mechanisms that rely on purely text-based rethinking processes perform well in most multimodal tasks. However, when directly applied

ROAD: Adaptive Data Mixing for Offline-to-Online Reinforcement Learning via Bi-Level Optimization

arXiv:2605.14497v1 Announce Type: cross Abstract: Offline-to-online reinforcement learning harnesses the stability of offline pretraining and the flexibility of online fine-tuning. A key challenge lie

Road Maps as Free Geometric Priors: Weather-Invariant Drone Geo-Localization with GeoFuse

arXiv:2605.14925v1 Announce Type: new Abstract: Drone-view geo-localization aims to match a query drone image, often captured under adverse weather conditions (e.g., rain, snow, fog), against a galler

Safety-Constrained Reinforcement Learning with Post-Training Reachability Verification for Robot Navigation

arXiv:2605.14174v1 Announce Type: new Abstract: Safe navigation for mobile robots demands policies that remain reliable under the high-consequence perception uncertainty of cluttered environments. Yet

safetyarxiv-cs-ro

Second-Order Actor-Critic Methods for Discounted MDPs via Policy Hessian Decomposition

arXiv:2605.14982v1 Announce Type: cross Abstract: We address the discounted reward setting in reinforcement learning (RL). To mitigate the value approximation challenges in policy gradient methods, ac

Selective Safety Steering via Value-Filtered Decoding

arXiv:2605.14746v1 Announce Type: new Abstract: While large language models (LLMs) are trained to align with human values, their generations may still violate safety constraints. A growing line of wor

Self-Distilled Agentic Reinforcement Learning

arXiv:2605.15155v1 Announce Type: cross Abstract: Reinforcement learning (RL) has emerged as a central paradigm for post-training LLM agents, yet its trajectory-level reward signal provides only coars

Send the arXiv AI-generated slop, get a yearlong vacation from submissions

ArXiv will ban authors for one year if they submit papers containing obviously AI-generated content , with examples including hallucinated citations, placeholder text, or chatbot meta-comments left in

safetyars-technica

SimPersona: Learning Discrete Buyer Personas from Raw Clickstreams for Grounded E-Commerce Agents

arXiv:2605.14205v1 Announce Type: new Abstract: LLM-based web agents can navigate live storefronts, yet they often collapse to a single 'average buyer' policy, failing to capture the heterogeneous and

SkillFlow: Flow-Driven Recursive Skill Evolution for Agentic Orchestration

arXiv:2605.14089v1 Announce Type: new Abstract: In recent years, a variety of powerful LLM-based agentic systems have been applied to automate complex tasks through task orchestration. However, existi

Slot-MPC: Goal-Conditioned Model Predictive Control with Object-Centric Representations

arXiv:2605.14937v1 Announce Type: cross Abstract: Predictive world models enable agents to model scene dynamics and reason about the consequences of their actions. Inspired by human perception, object

Smooth Multi-Policy Causal Effect Estimation in Longitudinal Settings

arXiv:2605.14284v1 Announce Type: new Abstract: Comparative evaluation of multiple dynamic treatment policies is essential for healthcare and policy decisions, yet conventional longitudinal causal inf

SOCC-ICP: Semantics-Assisted Odometry based on Occupancy Grids and ICP

arXiv:2605.15074v1 Announce Type: new Abstract: Reliable pose estimation in previously unseen environments is a fundamental capability of autonomous systems. Existing LiDAR odometry methods typically

safetyarxiv-cs-ro