safety

Improving Labeling Consistency with Detailed Constitutional Definitions and AI-Driven Evaluation

arXiv:2605.24247v1 Announce Type: cross Abstract: Many automated labeling pipelines classify inputs into categories defined by a written specification, content moderation being a prominent use case. S

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Safety

Improving the Completeness and Comparability of Segment Disclosures: A Large Language Model Approach

DGX agent

arXiv:2605.23924v1 Announce Type: new Abstract: Segment-level disclosures are a central component of financial reporting, providing insight into firms' internal organization and the allocation of econ

safetyarxiv-cs-cl

26 May 2026

In our new paper, EDGE-OPD (EviDence GuidEd On-Policy Distillation), we introduce two key ideas for improving On-Policy Distillation (OPD). …

In our new paper, EDGE-OPD (EviDence GuidEd On-Policy Distillation), we introduce two key ideas for improving On-Policy Distillation (OPD). First, we use guided rollouts that inject privileged context

safetyemad-mostaque--x

Inference-Time Alignment of Diffusion Models via Trust-Region Iterative Twisted Sequential Monte Carlo

arXiv:2605.25123v1 Announce Type: cross Abstract: We study inference-time alignment for diffusion-based generative models, aiming to steer a base model toward high-reward outputs without updating its

Internalizing Outcome Supervision into Process Supervision: A New Paradigm for Reinforcement Learning for Reasoning

arXiv:2605.05226v2 Announce Type: replace-cross Abstract: The central challenge of reinforcement learning for reasoning lies not only in the sparsity of outcome-level supervision, but more fundamental

Inverting the Shield: Systematically Generating Safety Tests from Policy Specifications

arXiv:2605.24883v1 Announce Type: new Abstract: The widespread integration of Large Language Models (LLMs) necessitates rigorous and systematic safety evaluation. Existing paradigms either rely on con

Investigating the Interplay between Contextual and Parametric Chain-of-Thought Faithfulness under Optimization

arXiv:2605.24960v1 Announce Type: cross Abstract: Chain-of-Thought (CoT) faithfulness, i.e., whether CoTs genuinely reflect large language models' (LLM) underlying behavior, is typically evaluated und

Is Decentralized AI Governable? From Regulative Policy to Constitutive Protocol

arXiv:2605.24538v1 Announce Type: cross Abstract: Every major framework for governing artificial intelligence presupposes an identifiable entity -- a developer, deployer, or operator -- who can be hel

Is GPT-4o mini Blinded by its Own Safety Filters? Exposing the Multimodal-to-Unimodal Bottleneck in Hate Speech Detection

arXiv:2509.13608v2 Announce Type: replace Abstract: As Large Multimodal Models (LMMs) become integral to daily digital life, understanding their safety architectures is a critical problem for AI Align

IsaacIPC: Coupling High-Fidelity Simulation and Realistic Rendering for Contact-Rich Robotic Systems

arXiv:2605.24339v1 Announce Type: new Abstract: We present IsaacIPC, a robotic simulation framework that couples GPU accelerated incremental potential contact (IPC) with IsaacSim/Lab. IsaacIPC maps si

Iterative Feature Space Optimization through Incremental Adaptive Evaluation

arXiv:2501.14889v2 Announce Type: replace Abstract: Iterative feature space optimization involves systematically evaluating and adjusting the feature space to improve downstream task performance. Howe

Iterative Refinement Neural Operators are Learned Fixed-Point Solvers: A Principled Approach to Spectral Bias Mitigation

arXiv:2605.24041v1 Announce Type: cross Abstract: Neural operators serve as fast, data-driven surrogates for scientific modeling but typically rely on a monolithic, single-pass inference procedure tha

It's a shame the Pope didn't ask Chris Olah what happened to his plan to give up to 10% of Anthropic to the authors of the work they train o…

It's a shame the Pope didn't ask Chris Olah what happened to his plan to give up to 10% of Anthropic to the authors of the work they train on. (Spoiler: it never happened, and the authors on whose wor

safetygary-marcus--x

IVR-R1: Refining Trajectories through Iterative Visual-Grounded Reasoning in Reinforcement Learning

arXiv:2605.23997v1 Announce Type: cross Abstract: Multimodal large language models via reinforcement learning (RL) have demonstrated remarkable capabilities in complex visual reasoning tasks, yet they

Jailbreak to Protect: Buffering and Reinforcing via Temporary Jailbreaking for Safe Fine-Tuning in Large Language Models

arXiv:2605.24550v1 Announce Type: new Abstract: Fine-tuning-as-a-Service (FaaS) enables personalization of large language models (LLMs), but it can weaken safety-alignment under harmful fine-tuning at

Joint Optimization of Training and Inference in Federated Edge Learning via Constrained Multi-Objective Deep Reinforcement Learning

arXiv:2605.25916v1 Announce Type: new Abstract: Federated edge learning (FEEL) has recently emerged as a promising paradigm for achieving edge intelligence (EI) via enabling collaborative model traini

JT-SAFE-V2: Safety-by-Design Foundation Model with World-Context Data

arXiv:2605.24414v1 Announce Type: new Abstract: We introduce JT-Safe-V2, a large language model designed to advance the safety and trustworthiness of foundation models, extending our previous JT-Safe

KYA: A Framework-Agnostic Trust Layer for Autonomous Systems with Verifiable Provenance and Hierarchical Policy Composition

arXiv:2605.25376v1 Announce Type: cross Abstract: Observability tells operators when an agent is slow. KYA tells operators when an agent is wrong, drifting, leaking, or quietly going rogue. We present

Label-NTK Alignments and A Tighter Convergence Bound in the NTK Regime

arXiv:2605.25275v1 Announce Type: new Abstract: The Neural Tangent Kernel (NTK) framework explains optimization in over-parameterized neural networks via approximately linearized dynamics, yielding ex

Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation

arXiv:2605.25036v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) extend large language models with visual understanding, but remain vulnerable to hallucination, where outputs are

LAPLEX: The FFT of Learnable Laplace Kernels

arXiv:2605.24584v1 Announce Type: cross Abstract: Fast linear algebra in deep learning usually comes with a choice: fixed geometry and exact computation, as in the Fourier transform, or adaptive geome

Latent Representation Alignment for Offline Goal-Conditioned Reinforcement Learning

arXiv:2605.25740v1 Announce Type: new Abstract: Offline goal-conditioned reinforcement learning (GCRL) provides a practical framework for obtaining goal-reaching policies from fixed datasets. However,

LC-ERD: Mining Latent Logic for Self-Evolving Reasoning via Consistency-Regulated Reward Decomposition

arXiv:2605.24005v1 Announce Type: new Abstract: The evolution of Large Language Model (LLM) reasoning is bottlenecked by the scarcity of high-quality process data. While self-alignment via endogenous

Learning Concepts, Not Tokens: Self-Supervised Semantic Alignment for Language Models

arXiv:2603.29123v2 Announce Type: replace Abstract: The next-token prediction (NTP) objective trains language models to predict a single token at each step, even though many continuations can express

Learning High-Frequency Continuous Action Chunks in Latent Space

arXiv:2605.24931v1 Announce Type: new Abstract: Modern robotic policies increasingly rely on action chunking to execute complex tasks in the physical world. While action chunking improves temporal con

Learning in Low-Dimensional Subspaces: Orthogonal Bottlenecks for Reinforcement Learning

arXiv:2605.26012v1 Announce Type: cross Abstract: Deep reinforcement learning (RL) agents commonly rely on high-dimensional neural representations, despite growing evidence that task-relevant value an

Learning to Route Languages for Multilingual Policy Optimization

arXiv:2605.25360v1 Announce Type: new Abstract: Large language models~(LLMs) are trained on heterogeneous multilingual corpora, yet existing policy optimization methods often implicitly restrict each

LipoAgent: Coordinating Fine-Tuned LLM Agents for Safer Lipid Design

arXiv:2605.25250v1 Announce Type: new Abstract: Lipid nanoparticles (LNPs) are among the most clinically mature platforms for nucleic acid delivery, yet designing lipids that are both effective and bi

Locality Matters for Training-Free Audio Token Compression in Audio-Language Models

arXiv:2605.25179v1 Announce Type: new Abstract: Audio-language models (ALMs) are increasingly used for audio captioning, question answering, and open-ended audio understanding, but their inference cos

lol. OpenAI as the WeWork of AI. Literally called it, in those exact words, @CNBC w @carlquintanilla, 2024. Now even SoftBank is worried.

lol. OpenAI as the WeWork of AI. Literally called it, in those exact words, @CNBC w @carlquintanilla, 2024. Now even SoftBank is worried. SoftBank's own executives think Sam Altman is scamming their C

safetygary-marcus--x

Looking forward to speaking at BloombergTech next week with @shiringhaffary to discuss AI safety and our research progress at @LawZero_!

Looking forward to speaking at BloombergTech next week with @shiringhaffary to discuss AI safety and our research progress at @LawZero_! Is AI development progressing too quickly? @business' @shiringh

safetyyoshua-bengio--x

Machine Psychometrics: A Mathematical Psychology of Artificial Intelligence

arXiv:2605.23952v1 Announce Type: new Abstract: Artificial agents now generate behavior rich enough to invite trust, surprise, and concern, yet our evaluation tools still privilege capability scores o

MAGIC: Multimodal Alignment & Grounding-aware Instruction Coreset for Vision-Language Models

arXiv:2605.26004v1 Announce Type: cross Abstract: Instruction tuning of large vision-language models (LVLMs) increasingly depends on massive multimodal corpora, yet these datasets contain samples with

MAPLE: Multi-State Aggregated Policy Evaluation for AlphaZero in Imperfect-Information Games

arXiv:2605.24139v1 Announce Type: new Abstract: Imperfect-information games (IIGs) are challenging, as players must make decisions without fully observing the true game state. While AlphaZero has achi

MARS: Margin and Semantic-Aware Data Augmentation for Reward Modeling

arXiv:2602.17658v2 Announce Type: replace-cross Abstract: Reward modeling is central to alignment pipelines such as RLHF, RLAIF, and PPO-based policy optimization, yet its reliability is constrained b

MATO: Multi-objective Personalized Alignment with Test-time Optimization for Large Language Models

arXiv:2605.25342v1 Announce Type: new Abstract: Aligning large language models (LLMs) with diverse and multifaceted user preferences is a fundamental challenge in personalized AI systems. Existing mul

Measuring the Depth of LLM Unlearning via Activation Patching

arXiv:2605.24614v1 Announce Type: cross Abstract: Large language model (LLM) unlearning has emerged as a crucial post-hoc mechanism for privacy protection and AI safety, yet auditing whether target kn

MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents

arXiv:2602.02474v2 Announce Type: replace-cross Abstract: Most Large Language Model (LLM) agent memory systems rely on a small set of static, hand-designed operations for extracting memory. These fixe

Metacognition Should Be the Scientific Framework for Bounded and Effective Self-Governance in Generative AI

arXiv:2605.23981v1 Announce Type: cross Abstract: Generative AI research increasingly confronts a shared problem: systems must sustain yet govern their own generative activity when uncertainty is high

Micro-Swarm Locomotion Optimization in Dynamic Flow using Multi-Objective Multi-Agent Reinforcement Learning

arXiv:2605.25025v1 Announce Type: new Abstract: Coordinating micro-robotic swarms in physiologically realistic, time-dependent fluid environments remains an unsolved challenge for biomedical and envir

Mitigating Hallucinations in Healthcare LLMs with Granular Fact-Checking and Domain-Specific Adaptation

arXiv:2512.16189v3 Announce Type: replace Abstract: In healthcare, it is essential for any LLM-generated output to be reliable and accurate, particularly in cases involving decision-making and patient

MMUEChange: A Generalized LLM Agent Framework for Intelligent Multi-Modal Urban Environment Change Analysis

arXiv:2601.05483v2 Announce Type: replace Abstract: Understanding urban environment change is essential for sustainable development. However, current approaches, particularly remote sensing change det

Motion-Compensated Weight Compression

arXiv:2605.24754v1 Announce Type: cross Abstract: Neural network weights are increasingly a bottleneck for deployment, yet most compression pipelines treat layers independently and overlook cross-laye

MuGen: Multi-Skill Generative Locomotion Controller for Humanoid Robots

arXiv:2605.24592v1 Announce Type: new Abstract: This paper presents MuGen, a data-driven framework for learning and deploying multi-skill locomotion on humanoid robots. MuGen enables a robot to perfor

Multi-Agent Coordination Adaptation via Structure-Guided Orchestration

arXiv:2605.25746v1 Announce Type: cross Abstract: As large language model (LLM)-based multi-agent systems scale to handle increasingly complex tasks, balancing structural stability and dynamic adaptab

Multi-Agent Systems are Mixtures of Experts: Who Becomes an Influencer?

arXiv:2605.25929v1 Announce Type: cross Abstract: The effectiveness of multi-agent LLM deliberation depends not only on the agents' individual predictions, but also on how they communicate and collabo

Multi-Alignment Contrastive Learning for Enzyme--Reaction Retrieval

arXiv:2512.08508v2 Announce Type: replace-cross Abstract: Identifying enzymes that catalyze target biochemical reactions is a key step in computational enzyme discovery and biocatalyst design. Recent

Multi-Modal Cross-Domain Alignment Network for Video Moment Retrieval

arXiv:2209.11572v3 Announce Type: replace-cross Abstract: As an increasingly popular task in multimedia information retrieval, video moment retrieval (VMR) aims to localize the target moment from an u