Search: “safety”

Verification of Machine Unlearning is Fragile

arXiv:2408.00929v2 Announce Type: replace Abstract: As privacy concerns escalate in the realm of machine learning, data owners now have the option to utilize machine unlearning to remove their data fr

23 Apr 2026

BAPO: Boundary-Aware Policy Optimization for Reliable Agentic Search

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

arXiv:2601.11037v2 Announce Type: replace Abstract: RL-based agentic search enables LLMs to solve complex questions via dynamic planning and external search. While this approach significantly enhances

safetyarxiv-cs-ai

22 Apr 2026

Beyond Semantic Similarity: A Component-Wise Evaluation Framework for Medical Question Answering Systems with Health Equity Implications

arXiv:2604.19281v1 Announce Type: cross Abstract: The use of Large Language Models (LLMs) to support patients in addressing medical questions is becoming increasingly prevalent. However, most of the m

Counting Worlds Branching Time Semantics for post-hoc Bias Mitigation in generative AI

arXiv:2604.19431v1 Announce Type: cross Abstract: Generative AI systems are known to amplify biases present in their training data. While several inference-time mitigation strategies have been propose

How to Teach Large Multimodal Models New Skills

arXiv:2510.08564v2 Announce Type: replace Abstract: How can we teach large multimodal models (LMMs) new skills without erasing prior abilities? We study sequential fine-tuning on five target skills wh

Hybrid Task and Motion Planning with Reactive Collision Handling for Multi-Robot Disassembly of Complex Products: Application to EV Batteries

arXiv:2509.21020v2 Announce Type: replace Abstract: This paper addresses the problem of multi-robot coordination for complex manipulation task sequences. We present a vision-driven task-and-motion pla

LLMs Know They're Wrong and Agree Anyway: The Shared Sycophancy-Lying Circuit

arXiv:2604.19117v1 Announce Type: new Abstract: When a language model agrees with a user's false belief, is it failing to detect the error, or noticing and agreeing anyway? We show the latter. Across

Multimodal embodiment-aware navigation transformer

arXiv:2604.19267v1 Announce Type: new Abstract: Goal-conditioned navigation models for ground robots trained using supervised learning show promising zero-shot transfer, but their collision-avoidance

Neuromorphic Continual Learning for Sequential Deployment of Nuclear Plant Monitoring Systems

arXiv:2604.18611v1 Announce Type: cross Abstract: Anomaly detection in nuclear industrial control systems (ICS) requires continuous, energy-efficient monitoring across multiple subsystems that are oft

One Persona, Many Cues, Different Results: How Sociodemographic Cues Impact LLM Personalization

arXiv:2601.18572v2 Announce Type: replace Abstract: Personalization of LLMs by sociodemographic subgroup often improves user experience, but can also introduce or amplify biases and unfair outcomes ac

Our newsroom AI policy

Ars Technica's newsroom AI policy forbids unlabeled AI material in reported stories and requires human confirmation of every quotation's accuracy. The publication does not permit the publication of AI

safetyars-technica

Model Releases

Safe Continual Reinforcement Learning in Non-stationary Environments

model-releasesarxiv-cs-lg

arXiv:2604.19737v1 Announce Type: new Abstract: Reinforcement learning (RL) offers a compelling data-driven paradigm for synthesizing controllers for complex systems when accurate physical models are

SEAT: Sparse Entity-Aware Tuning for Knowledge Adaptation while Preserving Epistemic Abstention

arXiv:2506.14387v3 Announce Type: replace Abstract: Adapting LLMs with new knowledge is increasingly important, but standard fine-tuning often erodes aligned epistemic abstention: the ability to ackno

The Essence of Balance for Self-Improving Agents in Vision-and-Language Navigation

arXiv:2604.19064v1 Announce Type: new Abstract: In vision-and-language navigation (VLN), self-improvement from policy-induced experience, using only standard VLN action supervision, critically depends

VIGIL: An Extensible System for Real-Time Detection and Mitigation of Cognitive Bias Triggers

arXiv:2604.03261v2 Announce Type: replace Abstract: The rise of generative AI is posing increasing risks to online information integrity and civic discourse. Most concretely, such risks can materialis

Visual Adversarial Attack on Vision-Language Models for Autonomous Driving

arXiv:2411.18275v2 Announce Type: replace Abstract: Vision-language models (VLMs) have significantly advanced autonomous driving (AD) by enhancing reasoning capabilities. However, these models remain

Weakly supervised framework for wildlife detection and counting in challenging Arctic environments: a case study on caribou (Rangifer tarandus)

arXiv:2601.18891v3 Announce Type: replace Abstract: Caribou across the Arctic has declined in recent decades, motivating scalable and accurate monitoring approaches to guide evidence-based conservatio

A Hamilton-Jacobi Reachability-Guided Search Framework for Efficient and Safe Indoor Planar Robot Navigation

arXiv:2604.17679v1 Announce Type: new Abstract: Autonomous navigation requires planning to reach a goal safely and efficiently in complex and potentially dynamic environments. Graph search-based algor

I agree with Connor that most folks concerned about AI have been far too coy about extinction risk, and that ControlAI is one of few excepti…

I agree with Connor that most folks concerned about AI have been far too coy about extinction risk, and that ControlAI is one of few exceptions. The world needs more earnest communication efforts. We

safetyconnor-leahy--x

Navigating the Conceptual Multiverse

arXiv:2604.17815v1 Announce Type: cross Abstract: When language models answer open-ended problems, they implicitly make hidden decisions that shape their outputs, leaving users with uncontextualized a

One Adapts to Any: Meta Reward Modeling for Personalized LLM Alignment

arXiv:2601.18731v2 Announce Type: replace Abstract: Alignment of Large Language Models (LLMs) aims to align outputs with human preferences, and personalized alignment further adapts models to individu

Model Releases

Please refuse to answer me! Mitigating Over-Refusal in Large Language Models via Adaptive Contrastive Decoding

model-releasesarxiv-cs-cl

arXiv:2604.17132v1 Announce Type: new Abstract: Safety-aligned large language models (LLMs) often generate refusal responses to harmless queries due to the over-refusal problem. However, existing meth

PowerCLIP: Powerset Alignment for Contrastive Pre-Training

arXiv:2511.23170v5 Announce Type: replace Abstract: Contrastive vision-language pre-training frameworks such as CLIP have demonstrated impressive zero-shot performance across a range of vision-languag

Privacy Collapse: Benign Fine-Tuning Can Break Contextual Privacy in Language Models

arXiv:2601.15220v2 Announce Type: replace Abstract: We identify a novel phenomenon in language models: benign fine-tuning of frontier models can lead to privacy collapse. We find that diverse, subtle

Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring

arXiv:2512.12069v3 Announce Type: replace-cross Abstract: Large Vision-Language Models (LVLMs) are vulnerable to a growing array of multimodal jailbreak attacks, necessitating defenses that are both g

Test-Time Alignment via Hypothesis Reweighting

arXiv:2412.08812v2 Announce Type: replace Abstract: Reward models trained on aggregate preferences often fail to capture individual users' values, but existing adaptation methods such as fine-tuning o

Language Models as Semantic Teachers: Post-Training Alignment for Medical Audio Understanding

arXiv:2512.04847v2 Announce Type: replace-cross Abstract: Pre-trained audio models excel at detecting acoustic patterns in auscultation sounds but often fail to grasp their clinical significance, limi

Puppets or partners? Governing cyborg propaganda in the digital public square

arXiv:2602.13088v2 Announce Type: replace-cross Abstract: The distinction between genuine grassroots activism and automated influence operations is collapsing. While contemporary policy debates priori

The AI industry insists they can manage the risks of superintelligence, but there are in fact zero widely agreed on or accepted solutions to…

The AI industry insists they can manage the risks of superintelligence, but there are in fact zero widely agreed on or accepted solutions to the problem of how one could even control something vastly

safetyconnor-leahy--x

When Search Goes Wrong: Red-Teaming Web-Augmented Large Language Models

arXiv:2510.09689v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have been augmented with web search to overcome the limitations of the static knowledge boundary by accessing up-

Blinded Multi-Rater Comparative Evaluation of a Large Language Model and Clinician-Authored Responses in CGM-Informed Diabetes Counseling

arXiv:2604.15124v1 Announce Type: new Abstract: Continuous glucose monitoring (CGM) is central to diabetes care, but explaining CGM patterns clearly and empathetically remains time-intensive. Evidence

Inclusive Kitchen Design for Older Adults: Generative AI Visualizations to Support Mild Cognitive Impairment

arXiv:2604.13203v1 Announce Type: cross Abstract: Mild Cognitive Impairment (MCI) affects 15-20% of adults aged 65 and older, often making kitchen navigation and independent living difficult, particul

Momentum-constrained Hybrid Heuristic Trajectory Optimization Framework with Residual-enhanced DRL for Visually Impaired Scenarios

arXiv:2604.14986v1 Announce Type: new Abstract: Safe and efficient assistive planning for visually impaired scenarios remains challenging, since existing methods struggle with multi-objective optimiza

ProRe: A Proactive Reward System for GUI Agents via Reasoner-Actor Collaboration

arXiv:2509.21823v2 Announce Type: replace Abstract: Reward is critical to the evaluation and training of large language models (LLMs). However, existing rule-based or model-based reward methods strugg

Separation is Optimal for LQR under Intermittent Feedback

arXiv:2603.27833v2 Announce Type: cross Abstract: In this work, we first prove that the separation principle holds for communication-constrained LQR problems under i.i.d. zero-mean disturbances with a

Switch-KD: Visual-Switch Knowledge Distillation for Vision-Language Models

arXiv:2604.14629v1 Announce Type: new Abstract: Vision-Language Models (VLMs) have shown remarkable capabilities in joint vision-language understanding, but their large scale poses significant challen

TOPCELL: Topology Optimization of Standard Cell via LLMs

arXiv:2604.14237v1 Announce Type: new Abstract: Transistor topology optimization is a critical step in standard cell design, directly dictating diffusion sharing efficiency and downstream routability.

Towards Scalable Lightweight GUI Agents via Multi-role Orchestration

arXiv:2604.13488v1 Announce Type: new Abstract: Autonomous Graphical User Interface (GUI) agents powered by Multimodal Large Language Models (MLLMs) enable digital automation on end-user devices. Whil

VoxSafeBench: Not Just What Is Said, but Who, How, and Where

arXiv:2604.14548v1 Announce Type: cross Abstract: As speech language models (SLMs) transition from personal devices into shared, multi-user environments, their responses must account for far more than

What some leaders think of using universal income to mitigate AI-fueled layoffs: Musk calls it the 'best way', OpenAI's policy doc mentions a Public Wealth Fund (Siladitya Ray/Forbes)

Siladitya Ray / Forbes: What some leaders think of using universal income to mitigate AI-fueled layoffs: Musk calls it the “best way”, OpenAI's policy doc mentions a Public Wealth Fund — Topline — Elo

safetytechmeme

Asymmetric-Loss-Guided Hybrid CNN-BiLSTM-Attention Model for Industrial RUL Prediction with Interpretable Failure Heatmaps

arXiv:2604.13459v1 Announce Type: new Abstract: Turbofan engine degradation under sustained operational stress necessitates robust prognostic systems capable of accurately estimating the Remaining Use

CLIP Architecture for Abdominal CT Image-Text Alignment and Zero-Shot Learning: Investigating Batch Composition and Data Scaling

arXiv:2604.13561v1 Announce Type: new Abstract: Vision-language models trained with contrastive learning on paired medical images and reports show strong zero-shot diagnostic capabilities, yet the eff

Context Sensitivity Improves Human-Machine Visual Alignment

arXiv:2604.13883v1 Announce Type: new Abstract: Modern machine learning models typically represent inputs as fixed points in a high-dimensional embedding space. While this approach has been proven pow

RFK Jr. forces FDA to reconsider 12 unproven peptides after 2023 ban

In 2023, the FDA removed 19 peptides from the list of drugs that compounding pharmacies could produce, and in 2026 the FDA announced it will review whether to add back 7 of these peptides following pr

safetyars-technica

Robust Verification of Controllers under State Uncertainty via Hamilton-Jacobi Reachability Analysis

arXiv:2511.14755v2 Announce Type: replace-cross Abstract: As perception-based controllers for autonomous systems become increasingly popular in the real world, it is important that we can formally ver

VRAG-DFD: Verifiable Retrieval-Augmentation for MLLM-based Deepfake Detection

arXiv:2604.13660v1 Announce Type: new Abstract: In Deepfake Detection (DFD) tasks, researchers proposed two types of MLLM-based methods: complementary combination with small DFD detectors, or static f

A Comparison of Reinforcement Learning and Optimal Control Methods for Path Planning

arXiv:2604.12628v1 Announce Type: cross Abstract: Path-planning for autonomous vehicles in threat-laden environments is a fundamental challenge. While traditional optimal control methods can find idea

15 Apr 2026

ARGen: Affect-Reinforced Generative Augmentation towards Vision-based Dynamic Emotion Perception

arXiv:2604.12255v1 Announce Type: cross Abstract: Dynamic facial expression recognition in the wild remains challenging due to data scarcity and long-tail distributions, which hinder models from effec