Search: “safety”

BOKBO (Best of K Bad Options): Calibrated Abstention for VLA Policies

arXiv:2605.30660v1 Announce Type: new Abstract: Test-time scaling for vision-language-action (VLA) policies, methods such as RoboMonkey, SEAL, MG-Select, and V-GPS, samples K candidate action chunks a

Chain-of-Thought Reasoning In The Wild Is Not Always Faithful

DGX agent

arXiv:2503.08679v5 Announce Type: replace Abstract: Recent studies indicate that when faced with explicit biases in prompts, models often omit mentioning these biases in their Chain-of-Thought (CoT) o

model-releasesarxiv-cs-ai

1 Jun 2026

EMBGuard: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied Agents

arXiv:2605.30924v1 Announce Type: new Abstract: MLLM-powered embodied agents deployed in real-world environments encounter physical hazards. However, existing approaches lack explicit mechanisms for i

EUDAIMONIA: Evaluating Undesirable Dynamics in AI

arXiv:2605.30654v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as conversational partners for companionship, emotional disclosure, and interpersonal advice, but t

Mellum2 Technical Report

arXiv:2605.31268v1 Announce Type: new Abstract: We present Mellum 2, an open-weight 12B-parameter Mixture-of-Experts (MoE) language model with 2.5B active parameters per token. Mellum 2 is a general-p

Same Patient, Different Words, Different Diagnosis? Evaluating Semantic Stability in Clinical LLMs

arXiv:2605.30646v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly used in clinical applications. However, their behavior remains highly sensitive to subtle linguistic var

TAGA: A Tangent-Based Reactive Approach for Socially Compliant Robot Navigation Around Human Groups

model-releasesarxiv-cs-ro

arXiv:2503.21168v3 Announce Type: replace Abstract: Robots navigating human-populated environments must avoid collisions while respecting the social structure of crowds, particularly the implicit boun

Target-Agnostic Calibration under Distribution Shift with Frequency-Aware Gradient Rectification

arXiv:2508.19830v2 Announce Type: replace-cross Abstract: Real-world model deployments inevitably encounter distribution shifts, rendering the confidence estimates of deep neural networks highly unrel

Industry

The state of AI right now

Elon Musk shared observations about the current state of artificial intelligence development and capabilities on X (formerly Twitter). The post likely discusses recent AI advances, challenges, or Musk

industryelon-musk--x

When LLMs Learn to Be Consistently Wrong: A Multi-Model Study of Linear Representations of Synthetic Deception

arXiv:2605.30381v1 Announce Type: cross Abstract: Deceptive alignment, in which models maintain accurate internal representations while deliberately producing false outputs, remains a central challeng

Tutorials

A shared playbook for trustworthy third party evaluations

This document outlines OpenAI's framework and recommendations for conducting independent third-party evaluations of AI systems to ensure trustworthiness and accountability. It establishes shared stand

tutorialsopenai

Tools

Auto-review mode is now available in Cursor. It allows agents to run tool calls with fewer approval prompts and safer execution.

Cursor has introduced an auto-review mode feature that enables AI agents to execute tool calls with reduced approval requirements while maintaining safer execution practices. This feature streamlines

toolscursor--x

ChatGPT diagnosed 40 million people with a disease that was invented as a joke. Not a real disease. Not a misunderstood disease. A completel…

model-releasesgary-marcus--x

ChatGPT diagnosed 40 million people with a disease that was invented as a joke. Not a real disease. Not a misunderstood disease. A completely fictional condition with a fake name, fake papers, and fak

Industry

DOJ sues states that rejected ICE requests for undercover license plates

The Department of Justice filed lawsuits against Maine, Massachusetts, Oregon, and Washington state alleging their refusal to issue undercover license plates to federal agents imposes unconstitutional

industryars-technica

Feature Geometry of LoRA Adapters: A Sparse Autoencoder Analysis of Representational Divergence in Fine-Tuned Language Models

arXiv:2605.28896v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has emerged as a widely adopted approach for adapting large language models, yet the internal representational changes induce

FinGuard: Detecting Financial Regulatory Non-Compliance in LLM Interactions

arXiv:2605.29427v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly deployed in financial services, a single non-compliant interaction can expose institutions to regulator

GPIC: A Giant Permissive Image Corpus for Visual Generation

arXiv:2605.30341v1 Announce Type: cross Abstract: Studying scalable methods for visual generative modeling requires large, accessible, and stable datasets. We introduce GPIC, a Giant Permissive Image

How Reliable Are AI Attackers Against a Fixed Vulnerable Target? A 400-Run Empirical Study of LLM Penetration Testing Consistency

arXiv:2605.30096v1 Announce Type: cross Abstract: Large language models (LLMs) can autonomously conduct multi-stage cyber attacks, but the consistency of their offensive behavior under repeated trials

Latent Performance Profiling of Large Language Models

arXiv:2605.30018v1 Announce Type: new Abstract: Large language models (LLMs) frequently achieve impressive scores on standardized benchmarks, yet accuracy alone offers a limited view of their capabili

MechELK: A Mechanistic Interpretability Framework for Eliciting Latent Knowledge in Large Language Models

arXiv:2605.28825v1 Announce Type: new Abstract: Large language models (LLMs) frequently encode factual and reasoning knowledge in their internal representations that is not faithfully reflected in the

NICE: A Theory-Grounded Diagnostic Benchmark for Social Intelligence of LLMs

arXiv:2605.29685v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly applied in social contexts such as emotional companionship and customer service, measuring their social

Recovering Diversity Without Losing Alignment: A DPO Recipe for Post-Trained LLMs

arXiv:2605.30021v1 Announce Type: new Abstract: Many open-ended instructions have multiple valid answers that users can benefit from seeing, but post-training often narrows an LLM's output space towar

Agents

self-modification

Self-modification in AI systems refers to the capability of an artificial intelligence to alter its own code, parameters, or behavior patterns without external intervention. This concept, discussed by

agentsyohei-nakajima--x

Steering at the Source: Style Modulation Heads for Robust Persona Control

arXiv:2603.13249v2 Announce Type: replace-cross Abstract: Activation steering offers a computationally efficient mechanism for controlling Large Language Models (LLMs) without fine-tuning. While effec

SURGENT: A Surgical Multi-Agent Assistance System Across the Perioperative Workflow

arXiv:2605.29368v1 Announce Type: cross Abstract: The intricate nature of modern surgical care necessitates intelligent systems that can synthesize extensive patient records, support collaborative dec

TANDEM: Temporal-Aware Neural Detection for Multimodal Hate Speech

arXiv:2601.11178v2 Announce Type: replace Abstract: Social media platforms are increasingly dominated by long-form multimodal content, where harmful narratives are constructed through a complex interp

AI in SRE: Where and how Google is deploying agentic AI to improve operations

model-releasesgoogle-cloud-ai

Since its inception over 20 years ago, Google has used Site Reliability Engineering (SRE) to keep services like Search, Gmail, Maps, YouTube and Google Cloud reliable and highly available, adhering to

Debate with Images: Detecting Deceptive Behaviors in Multimodal Large Language Models

arXiv:2512.00349v2 Announce Type: replace Abstract: Are frontier AI systems becoming more capable? Certainly. Yet such progress is not an unalloyed blessing but rather a Trojan horse: behind their per

Localizing Input Uncertainty Quantification for Large Language Models via Shapley Values

arXiv:2605.28170v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly integrated into high-stakes decision-making, the ability to reliably quantify uncertainty has become a

Multi-Adapter Representation Interventions via Energy Calibration

arXiv:2605.28722v1 Announce Type: new Abstract: Representation intervention has emerged as a promising paradigm for aligning large language models toward desired behaviors without modifying model weig

PAST2HARM: A Simple Adaptive Past Tense Attack for Jailbreaking Multimodal AI

arXiv:2605.27545v1 Announce Type: new Abstract: Jailbreak attacks on multimodal AI systems remain underexplored, even though unsafe image generation can have more severe consequences than unsafe text

Persuade Me if You Can: A Framework for Evaluating Persuasion Effectiveness and Susceptibility Among Large Language Models

arXiv:2503.01829v4 Announce Type: replace-cross Abstract: Large Language Models (LLMs) demonstrate persuasive capabilities that rival human-level persuasion. While these capabilities can be used for s

Edge AI Deployment Beyond Models: A BSP-Aware Systems Framework for Industrial Embedded Platforms

arXiv:2605.26119v1 Announce Type: cross Abstract: Industrial Edge AI programs often begin with the model and only later confront the platform. That sequencing is attractive because it allows early dem

On the Sensitivity of Instruction-tuned LLMs to Harmful Sentences in Long Inputs

arXiv:2510.05864v2 Announce Type: replace Abstract: Large language models (LLMs) increasingly operate on long inputs, yet their behavior when harmful sentences are sparsely embedded within such inputs

Persistent AI Agents in Academic Research: A Single-Investigator Implementation Case Study

arXiv:2605.26870v1 Announce Type: cross Abstract: Background: Large language models are typically evaluated as models, benchmarks, or short conversational episodes. Less is known about what happens wh

SEC-bench Pro: Can Language Models Solve Long-Horizon Software Security Tasks?

arXiv:2605.26548v1 Announce Type: cross Abstract: Large language models (LLMs) now support automated software security tasks, including vulnerability discovery and proof-of-concept (PoC) generation. E

Sentinel: Embodied Cooperative Spatial Reasoning and Planning

model-releasesarxiv-cs-cv

arXiv:2605.26239v1 Announce Type: new Abstract: In this work, we study Cooperative Spatial Intelligence, the ability of decentralized embodied agents to coordinate effectively under dynamic environmen

What Molecular Structure Cannot Tell Us: A Taxonomy of Explainability Gaps in GNN-Based Drug Toxicity Prediction

arXiv:2605.26183v1 Announce Type: cross Abstract: Graph Neural Networks (GNNs) have emerged as a structurally natural approach for molecular toxicity prediction, operating directly on atomic connectiv

AI Content Moderation in Therapy Conversations

arXiv:2605.25454v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly being used for emotional support. They are also being developed for formal therapy purposes. However, LL

Benchmarking and Learning Real-World Customer Service Dialogue

arXiv:2510.22143v3 Announce Type: replace Abstract: Existing benchmarks and training pipelines for industrial intelligent customer service (ICS) remain misaligned with real-world dialogue requirements

Emotional intelligence in large language models is fragmented across perception, cognition, and interaction

arXiv:2605.24686v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly integrated into emotionally sensitive domains, the structural integrity of their emotional intelligence

MDIA: A Multi-Agent Diagnostic Intelligence Pipeline on HealthBench Professional

arXiv:2605.24699v1 Announce Type: new Abstract: Most reported gains on agentic-LLM clinical benchmarks are often attributed to prompt engineering, yet our results suggest that larger improvements can

PEDESTRIANQA: A Benchmark for Vision-Language Models on Pedestrian Intention and Trajectory Prediction

arXiv:2605.24562v1 Announce Type: cross Abstract: Pedestrian intention and trajectory prediction are critical for the safe deployment of autonomous driving systems, directly influencing navigation dec

Reward-free Alignment for Conflicting Objectives

arXiv:2602.02495v3 Announce Type: replace-cross Abstract: Direct alignment methods are increasingly used to align large language models (LLMs) with human preferences. However, many real-world alignmen

Security in the Fine-Tuning Lifecycle of Large Language Models: Threats, Defenses,Evaluation, and Future Directions

arXiv:2605.25073v1 Announce Type: cross Abstract: Background: Fine-tuning is central to adapting pre-trained Large Language Models (LLMs) to downstream tasks, but its reliance on training data, parame

SemanticZip: A Pilot Framework for Lossy Text Compression with LLMs as Semantic Decompressors

arXiv:2605.24541v1 Announce Type: cross Abstract: Text compression for large language model (LLM) systems is usually framed as token deletion, retrieval, summarization, or exact reconstruction. We stu

Understanding and Mitigating Premature Confidence for Better LLM Reasoning

arXiv:2605.24396v1 Announce Type: new Abstract: Long chains of thought (CoT) from current language models frequently contain logical gaps and unjustified leaps, limiting the gains from additional test

Apparently this clip is too spicy! So let's try it this way! Examples of Director with LTX 2.3 and a few different techniques.