safety

Who Grades the Grader? Co-Evolving Evaluation Metrics and Skills for Self-Improving LLM Agents

arXiv:2607.12790v1 Announce Type: new Abstract: Self-evolving agent systems improve by creating, revising, and retiring their own skills, but every such loop rests on a hidden assumption: a reliable e

15 Jul 2026

Safety

little by little, OpenAI’s storytelling is falling apart. my 2023 projection that they would someday be viewed as the WeWork of AI is lookin…

DGX agent

little by little, OpenAI’s storytelling is falling apart. my 2023 projection that they would someday be viewed as the WeWork of AI is looking stronger by the day. OpenAI is on pace to miss its own fiv

safetygary-marcus--x

14 Jul 2026

New work led by @FlemmingKondrup and @tomjiralerspong highlights an important vulnerability in chain-of-thought monitoring for agents, I hig…

New work led by @FlemmingKondrup and @tomjiralerspong highlights an important vulnerability in chain-of-thought monitoring for agents, I highly recommend giving it a read. Link to the paper: https://a

safetyyoshua-bengio--x

14 Jul 2026

ScienceSoft’s HIPAA-compliant AI voice scheduler built on AWS

In this post, you will learn how ScienceSoft, an Amazon Web Services (AWS) Services Partner, integrated Amazon Nova 2 Sonic with Amazon Bedrock Guardrails to build a Health Insurance Portability and A

safetyaws-ml-blog

14 Jul 2026

Absolutely fascinating work by @SakanaAILabs reproducing @kenneth0stanley Picbreeder in a non-interactive, VLM-agentic way. I've had years t…

Absolutely fascinating work by @SakanaAILabs reproducing @kenneth0stanley Picbreeder in a non-interactive, VLM-agentic way. I've had years to reflect on Kenneth Stanley's ideas as originally communica

safetydavid-ha--x

For almost two decades people like @YLeCun and @geoffreyhinton dumped on me for saying we need symbols in addition to deep learning. But tha…

For almost two decades people like @YLeCun and @geoffreyhinton dumped on me for saying we need symbols in addition to deep learning. But that’s exactly what loop engineering is: adding symbols to deep

safetygary-marcus--x

The Anglo-Scottish Enlightenment – the real antidote to Rousseau and Voltaire The French Enlightenment and the Anglo-Scottish Enlightenment …

The Anglo-Scottish Enlightenment – the real antidote to Rousseau and Voltaire The French Enlightenment and the Anglo-Scottish Enlightenment happened simultaneously, in the same century, reading the sa

safetyelon-musk--x

@theo What @GaryMarcus has been saying ... the engineering around LLMs matters even more than the LLMs themselves today

Gary Marcus argues that the engineering and infrastructure surrounding large language models are more critical to their practical success than the models themselves. This reflects his broader perspect

safetygary-marcus--x

A Collaborative Reasoning Framework for Anomaly Diagnostics in Underwater Robotics

arXiv:2511.03075v2 Announce Type: replace Abstract: The safe deployment of autonomous systems in safety-critical settings requires a paradigm that combines human expertise with AI-driven analysis, esp

safetyarxiv-cs-ro

A First-Principles Theory of Slow Thinking and Active Perception

arXiv:2607.08196v1 Announce Type: new Abstract: As part of a series on first-principles modeling of cognitive functions, this paper attempts to provide a mathematical formulation of thinking and perce

A US NLRB judge rules that Atlassian had illegally fired an employee in 2023 for pushing back against manager layoffs, and orders reinstatement and compensation (Noam Scheiber/New York Times)

Noam Scheiber / New York Times: A US NLRB judge rules that Atlassian had illegally fired an employee in 2023 for pushing back against manager layoffs, and orders reinstatement and compensation — A fed

safetytechmeme

ADORN: Adaptive Drift handling for Open RAN using Reinforcement Learning

arXiv:2607.08443v1 Announce Type: cross Abstract: Dynamic traffic variations in Open Radio Access Networks (O-RAN) lead to drift, which degrades the performance of Artificial Intelligence/Machine Lear

Ahead of a dinner with a US senator, AI researcher Nate Soares (@So8res) was told: 'Don't give them any of the crazy crap. You know, play it…

Ahead of a dinner with a US senator, AI researcher Nate Soares (@So8res) was told: 'Don't give them any of the crazy crap. You know, play it cool.' His friends opened with the concern that someone cou

safetyconnor-leahy--x

Aleena: Alignment Agent for Research Software Engineering Collaborations

arXiv:2607.08043v1 Announce Type: cross Abstract: Research software collaborations span meetings, informal chats, pull requests, and GitHub issues. A decision surfaced in a Slack thread, refined in a

Alignment Plausibility: A New Standard for Assuring AI in Healthcare

arXiv:2607.07766v1 Announce Type: new Abstract: Large language models (LLMs) have become significant providers of mental health support, yet they remain products of an attention economy whose operatio

As part of our ongoing efforts to strengthen our safeguards for advanced AI capabilities in biology, we’re evolving our Bio Bug Bounty into …

As part of our ongoing efforts to strengthen our safeguards for advanced AI capabilities in biology, we’re evolving our Bio Bug Bounty into an ongoing private program, known as the OpenAI Bio Bug Boun

safetyopenai--x

Bayesian Experimental Design via Score Matching

arXiv:2607.08335v1 Announce Type: cross Abstract: Policy-based approaches to Bayesian experimental design (BED) allow the learning of deep policy networks that adaptively make intelligent design decis

Best-of-N TTS Evaluation is Confounded by ASR Family Alignment

arXiv:2607.08256v1 Announce Type: cross Abstract: Best-of-N (BoN) inference improves content consistency in zero-shot text-to-speech by selecting from N candidates with an automatic speech recognition

Beyond Success Rates: Trainability and Extractability for Offline GCRL

arXiv:2602.05459v2 Announce Type: replace Abstract: Offline goal-conditioned reinforcement learning (GCRL) is typically benchmarked by the best tuned success rate of each method. This score measures a

Borrowing from anything: A generalizable framework for reference-guided instance editing

arXiv:2512.15138v2 Announce Type: replace Abstract: Reference-guided instance editing is fundamentally limited by semantic entanglement, where a reference's intrinsic appearance is intertwined with it

breaking: company built on stolen IP and lies allegedly steals more IP

breaking: company built on stolen IP and lies allegedly steals more IP “OpenAI’s nascent hardware business now rests on the shakiest of foundations, rotten to its core by its illegal reliance on misap

safetygary-marcus--x

Bridging Cognitive Neuroscience and Graph Intelligence: Hippocampus-Inspired Multi-View Hypergraph Learning for Web Finance Fraud

arXiv:2601.11073v3 Announce Type: replace-cross Abstract: Online financial services constitute an essential component of contemporary web ecosystems, yet their openness introduces substantial exposure

CAAD: Causality-Aware Multivariate Time Series Anomaly Detection via Multi-Scale Alignment and Structural Causal Consistency

arXiv:2607.08555v1 Announce Type: new Abstract: The operational integrity of complex industrial systems relies on precise anomaly detection and diagnosis. The vast majority of existing methods narrowl

ContactMimic: Humanoid Object Interaction via Contact Control

arXiv:2607.08742v1 Announce Type: new Abstract: Keypoint tracking alone is insufficient for object interaction tasks such as sitting on a chair, wiping a board, or pushing furniture, where the robot c

safetyarxiv-cs-ro

Contravariance Theory: Strong Alignment for Minimal Solutions to Hard Tasks

arXiv:2607.08561v1 Announce Type: new Abstract: A series of results from the NeuroAI over the past fifteen years have raised core questions both about how to compare Deep Neural Network (DNN) models t

Contributing to U.K. financial sector resilience as a critical third party

At Google Cloud, we take our role in the financial ecosystem very seriously. We firmly believe that operational resilience is essential to driving and sustaining responsible innovation. Today, we mark

safetygoogle-cloud-ai

Curriculum Learning for Efficient Chain-of-Thought Distillation via Structure-Aware Masking and GRPO

arXiv:2602.17686v4 Announce Type: replace-cross Abstract: Distilling Chain-of-Thought (CoT) reasoning from large language models into compact student models presents a fundamental challenge: teacher r

DeltaDeno: Zero-Shot Anomaly Generation via Delta-Denoising Attribution

arXiv:2511.16920v2 Announce Type: replace Abstract: Anomaly generation is often framed as few-shot fine-tuning with anomalous samples, which contradicts the scarcity that motivates generation and tend

Detecting Ladder Logic Bombs in IEC 61131-3 PLC Programs using ESBMC-PLC+: A Formal Verification Approach with Trigger Synthesis

arXiv:2607.08417v1 Announce Type: new Abstract: A Ladder Logic Bomb (LLB) is malicious control logic in a Programmable Logic Controller (PLC) program that lies dormant until a trigger activates a payl

Diagnosing Corruption-Induced Reliability Failures in Vision-Language Models

arXiv:2511.19032v2 Announce Type: replace Abstract: Visual corruptions can change vision--language model (VLM) behavior in ways that top-1 accuracy does not capture. A model may keep the same answer w

DKDNet: Dual Knowledge and Data-Driven Network for Cross-Domain Automatic Modulation Classification

arXiv:2607.08031v1 Announce Type: cross Abstract: The dynamics of communication environments induce significant distribution shifts across domains, challenging the generalization of deep learning-base

DR-Arena: an Automated Evaluation Framework for Deep Research Agents

arXiv:2601.10504v2 Announce Type: replace Abstract: As Large Language Models (LLMs) increasingly operate as Deep Research (DR) Agents capable of autonomous investigation and information synthesis, rel

DrugGen 2: A disease-aware language model for enhancing drug discovery

arXiv:2607.08404v1 Announce Type: cross Abstract: Current computational approaches for drug design typically focus on generating molecules conditioned on specific targets or general molecular properti

Dual-Difficulty Curriculum Learning for Direct Preference Optimization

arXiv:2504.07856v4 Announce Type: replace Abstract: Curriculum learning enhances Direct Preference Optimization (DPO) for aligning Large Language Models (LLMs), yet existing methods rely on a one-dime

Early to Share, Late to Save: Synchronisation-Driven Communication Gating in Bandwidth-Constrained Cooperative VLN

arXiv:2607.08504v1 Announce Type: cross Abstract: Most cooperative Vision-Language Navigation (VLN) methods assume unlimited communication, not considering real-world applications where bandwidth is r

safetyarxiv-cs-ro

Echoes: A semantically-aligned music deepfake detection dataset

arXiv:2603.23667v2 Announce Type: replace-cross Abstract: We introduce Echoes, a new dataset for music deepfake detection designed for training and benchmarking detectors under realistic and provider-

Efficient Partitioning Method of Large-Scale Public Safety Spatio-Temporal Data based on Information Loss Constraints

arXiv:2306.12857v3 Announce Type: replace Abstract: The storage, management, and application of massive spatio-temporal data are widely used in practical scenarios, including public safety. However, d

Efficient Safety Alignment of Language Models via Latent Personality Traits

arXiv:2607.07918v1 Announce Type: cross Abstract: Current safety methods for large language models are known to be vulnerable to adversarial attacks, motivating research into robust alternatives. Late

EgoWAM: World Action Models Beyond Pixels with In-the-Wild Egocentric Human Data

arXiv:2607.08436v1 Announce Type: cross Abstract: Egocentric human data offers scalable supervision for robot manipulation. However, behavior cloning entangles transferable content like objects, scene

Ensemble Diversity Optimization for Subjective Supervision

arXiv:2607.08493v1 Announce Type: cross Abstract: Subjective NLP tasks often exhibit systematic annotator disagreement, requiring models that represent uncertainty rather than collapse it. We introduc

EU finds that Meta breached bloc’s rules with its social network interfaces

The European Union has tentatively found that Meta Platforms Inc. breached the bloc’s DSA tech industry law. The European Commission, the EU’s executive arm, published its conclusions today. The DSA,

safetysiliconangle

Expressivity and Statistical Trade-offs in Diffusion Policy Learning

arXiv:2607.07967v1 Announce Type: cross Abstract: Diffusion-based policies have recently emerged as powerful policy parameterizations for reinforcement learning, representing state-conditioned action

Feedback Manipulation Regularization: Enabling Offline Agent Alignment for Imitation Learning

arXiv:2607.07859v1 Announce Type: new Abstract: Reinforcement learning (RL) research has increasingly shifted focus towards alignment, ensuring agents learn behaviors adhering to human values. While h

From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents

arXiv:2607.08028v1 Announce Type: new Abstract: Enterprise large language model (LLM) applications often begin as prototypes whose behavior is carried by prompts and retrieval context. Productization

Geometry-Aware Deep Congruence Networks for Manifold Learning in Cross-Subject Motor Imagery

arXiv:2511.18940v3 Announce Type: replace Abstract: Cross-subject motor imagery decoding remains a fundamental challenge in EEG-based brain-computer interfaces due to substantial inter-subject variabi

HairWeaver: Few-Shot Photorealistic Hair Motion Synthesis with Sim-to-Real Guided Video Diffusion

arXiv:2602.11117v2 Announce Type: replace Abstract: We present HairWeaver, a diffusion-based pipeline that animates a single human image with realistic and expressive hair dynamics. While existing met

HeaPA: Difficulty-Aware Heap Sampling and On-Policy Query Augmentation for LLM Reinforcement Learning

arXiv:2601.22448v2 Announce Type: replace-cross Abstract: RLVR has become a standard recipe for training LLMs on reasoning tasks with verifiable outcomes, but when rollout generation dominates the cos

HSA: Hierarchical Slot Attention for Multi-granularity Scene-Decomposition

arXiv:2607.08249v1 Announce Type: new Abstract: Slot attention is a powerful framework for object-centric learning, decomposing visual scenes into latent slots through iterative competitive attention.