03
← Back to the timeline
Daily archive

AI news on July 3, 2026 · Friday

100 stories — deduplicated across sources, ranked by significance, every source cited.
6
SIGNIFICANCE
★ Top story · Other8h ago

More details on Fable 5’s cyber safeguards and our jailbreak framework

Anthropic has detailed the safety protocols implemented in its latest Fable model, focusing on the defense mechanisms used to prevent unauthorized output and malicious prompting. The company also introduced a new testing framework designed to systematically identify and address jailbreak vulnerabilities. By publicizing these internal evaluation tools, Anthropic aims to provide developers with a clearer methodology for hardening large language models against adversarial attacks.

AAnthropicOpen full story →Read original ↗
4

Safeguarding LLM Agents from Misalignment through Provenance Analysis

Policy·arXiv CS.AI·4h agoA
4

The risk of KV cache compression

Policy·arXiv CS.AI·4h agoA
4

MultAttnAttrib: Training-Free Multimodal Attribution in Long Document Question Answering

Research·arXiv CS.AI·4h agoA
4

Teaching Vision-Language-Action Models What to See and Where to Look

Opinion·arXiv CS.AI·4h agoA
4

Geometric Signatures of Reasoning: A Spectral Perspective on Task Hardness

Opinion·arXiv CS.AI·4h agoA
4

ICDepth: Taming Video Diffusion Models for Video Depth Estimation via In-Context Conditioning

Research·arXiv CS.AI·4h agoA
4

Multi-Head Recurrent Memory Agents

Research·arXiv CS.AI·4h agoA
4

DeadPool: Resilient LLM Training with Hot-Swapping via Zero-Overhead Checkpoint

Models·arXiv CS.AI·4h agoA
4

Parameter Golf: What Really Works?

Opinion·arXiv CS.AI·4h agoA
4

Discrete Diffusion Language Models for Interactive Radiology Report Drafting

Open Source·arXiv CS.AI·4h agoA
4

C2E: Boosting Ego-Only 3D Object Detection via Multi-Teacher Contrastive Knowledge Distillation

Research·arXiv CS.AI·4h agoA
4

EPnG: Adaptive Expert Prune-and-Grow for Parameter-Efficient MoE Fine-tuning

Models·arXiv CS.AI·4h agoA
4

AnchorSplat: Fast and Structure Consistent Detail Synthesis for Gaussian Splatting

Research·arXiv CS.AI·4h agoA
4

Gaming Consensus: Coordinated Manipulation in Crowdsourced Fact-Checking

Research·arXiv CS.AI·4h agoA
4

Denser $\neq$ Better: Limits of On-Policy Self-Distillation for Continual Post-Training

Research·arXiv CS.AI·4h agoA
4

EO-Agents: A Three-Agent LLM Pipeline for Earth Observation Hypothesis Generation

Models·arXiv CS.AI·4h agoA
4

Janus: a Playground for User-Involved Agentic Permission Management

Research·arXiv CS.AI·4h agoA
4

kNNGuard: Turning LLM Hidden Activations into a Training-Free Configurable Guardrail

Research·arXiv CS.AI·4h agoA
4

Structure-Aware Gaussian Splatting for Large-Scale Scene Reconstruction

Research·arXiv CS.AI·4h agoA
4

IsoSci: A Benchmark of Isomorphic Cross-Domain Science Problems for Evaluating Reasoning versus Knowledge Retrieval in LLMs

Research·arXiv CS.AI·4h agoA
4

TokenScope: Token-Level Explainability and Interpretability for Code-Oriented Tasks in Large Language Models

Research·arXiv CS.AI·4h agoA
4

When Does Generating More Help? Disentangling Fixed-Source Synthesis from Source Expansion in Synthetic Data Scaling

Research·arXiv CS.AI·4h agoA
4

Autonomous discovery of traffic laws with AI traffic scientists

Policy·arXiv CS.AI·4h agoA
4

Beyond the Performance Illusion: Structure-Aware Stratified Partitioning and Curriculum Distributionally Robust Optimization for Spatially Correlated Domains

Research·arXiv CS.AI·4h agoA
4

Open-Weather Robust 3D Detection via Dual-Critic Diffusion Alignment

Policy·arXiv CS.AI·4h agoA
4

The Turning Point of 3D Plant Phenotyping: 3D Foundation Models Enable Minute-to-Second Cross-Crop Reconstruction and Beyond

Research·arXiv CS.AI·4h agoA
4

Safety Targeted Embedding Exploit via Refinement

Policy·arXiv CS.AI·4h agoA
4

Zeus: Towards Tuning-Free Foundation Model for Time Series Analysis

Industry·arXiv CS.AI·4h agoA
4

Black-Box Inference of LLM Architectural Properties with Restrictive API Access

Products·arXiv CS.AI·4h agoA
4

How to Allocate Your Tokens? Scaling Laws with Training Steps and Batch Size

Research·arXiv CS.AI·4h agoA
4

Conditional Co-Ablation: Recovering Self-Repair Backups in Transformer Circuits

Models·arXiv CS.AI·4h agoA
4

ContextSniper: AntTrail's Token-Efficient Code Memory for Repository-Level Program Repair

Open Source·arXiv CS.AI·4h agoA
4

Kara: Efficient Reasoning LLM Serving via Sliding-Window KV Cache Compression

Research·arXiv CS.AI·4h agoA
4

Path-level Hindsight Instructions for Semantic Exploration in Vision-Language Navigation

Models·arXiv CS.AI·4h agoA
4

Online Segment 3D Gaussians via Launching Virtual Drones

Models·arXiv CS.AI·4h agoA
4

Revisiting Chain-of-Thought Reasoning under Limited Supervision: Semi-supervised Chain-of-Thought Learning

Research·arXiv CS.AI·4h agoA
4

Scaling with Confidence: Calibrating Confidence of LLMs for Adaptive Test Time Scaling

Research·arXiv CS.AI·4h agoA
4

VLAFlow: A Unified Training Framework for Vision-Language-Action Models via Co-training and Future Latent Alignment

Research·arXiv CS.AI·4h agoA
4

Beyond Heatmaps: Unsupervised Concept-Graph Reasoning for Interpretable Visual Explanation

Research·arXiv CS.AI·4h agoA
4

Auto-FL-Research: Agentic Search for Federated Learning Algorithms

Research·arXiv CS.AI·4h agoA
4

Office Comprehension Benchmark

Research·arXiv CS.AI·4h agoA
4

PixGS: Pixel-Space Diffusion for Direct 3D Gaussian Splat Generation

Research·arXiv CS.AI·4h agoA
4

PACE: A Neuro-Symbolic Framework for Plausible and Actionable Counterfactual Explanations

Research·arXiv CS.AI·4h agoA
4

ElephantAgent: Contextual State Continuity in Agentic Systems

Research·arXiv CS.AI·4h agoA
4

MapDreamer: Aerial Imagery Conditioned Latent Diffusion for Lane-Level Map Generation

Research·arXiv CS.AI·4h agoA
4

Spec-AUF: Accept-Until-Fail Training under Train-Inference Misalignment for Masked Block Drafters

Research·arXiv CS.AI·4h agoA
4

HandsOnWorld: Unconstrained Egocentric Video Generation with Camera-Disentangled Hand Control

Models·arXiv CS.AI·4h agoA
4

When Should Service Agents Reconsider? Difficulty-Routed Control in Customer-Service Operations

Products·arXiv CS.AI·4h agoA
4

Object Aligner: A Configurable JSON Schema Similarity Score for Graphs, Applied to LLM Prompt Optimization

Products·arXiv CS.AI·4h agoA
4

Rethinking Complexity Metrics for LLM-Integrated Applications: Beyond Source Code

Products·arXiv CS.AI·4h agoA
4

MedStreamBench: A Time-Aware Benchmark for Streaming and Proactive Medical Video Understanding

Research·arXiv CS.AI·4h agoA
4

FaithMed: Training LLMs For Faithful Evidence-Based Medical Reasoning

Research·arXiv CS.AI·4h agoA
4

Epistemic Goggles: A Pretrained Module that Induces an Epistemic Frame via Gradient Editing

Research·arXiv CS.AI·4h agoA
4

Safe and Adaptive Cloud Healing: Verifying LLM-Generated Recovery Plans with a Neural-Symbolic World Model

Research·arXiv CS.AI·4h agoA
4

OpenSafeIntent: Evaluating Intent-Calibrated Safe Completion Across Dual-Use Prompt Sets

Research·arXiv CS.AI·4h agoA
4

Anti-Prompt: Image Protection against Text-Guided Image-to-Video Generation

Research·arXiv CS.AI·4h agoA
4

The Rollout Infrastructure Tax in Coding-Agent Reinforcement Learning

Research·arXiv CS.AI·4h agoA
4

Rank-Then-Act: Reward-Free Control from Frame-Order Progress

Research·arXiv CS.AI·4h agoA
4

Breaking Safety at the Token Boundary: How BPE Tokenization Creates Exploitable Gaps in LLM Alignment

Policy·arXiv CS.AI·4h agoA
4

Hierarchical Anti-Aesthetics: Protecting Facial Privacy against Customized Diffusion Models

Research·arXiv CS.AI·4h agoA
4

Do LLMs Truly Generalize in the Molecular Domain? A Perturbation-Based Analysis

Opinion·arXiv CS.AI·4h agoA
4

Spin-Weighted Spherical Harmonics Enable Complete and Scalable $\mathrm{E}(3)$-Equivariant Networks

Research·arXiv CS.AI·4h agoA
4

InduceKV: Fixed-Footprint Continual Adaptation of Multimodal LLMs via Inducing KV Memories

Models·arXiv CS.AI·4h agoA
4

Scaling Laws for Grid-Based Approximate Nearest Neighbor Search in High Dimensions

Products·arXiv CS.AI·4h agoA
4

Don't Let Gains FADE: Breaking Down Policy Gradient Weights in RL

Models·arXiv CS.AI·4h agoA
4

Hidden Forgetting in Continual Multimodal Learning: When Accuracy Survives but Grounding Fails

Models·arXiv CS.AI·4h agoA
4

Multi-THuMBS: Multi-person Tracking of 3D Human Meshes Beyond Video Shots

Research·arXiv CS.AI·4h agoA
4

Can Language Models Actually Retrieve In-Context? Drowning in Documents at Million Token Scale

Industry·arXiv CS.AI·4h agoA
4

DiPS: Dialogue Policy Selection for High-Stakes Persuasion Agents

Policy·arXiv CS.AI·4h agoA
4

X-LogSMask: Expand Transformer for Graph-Structured Data

Models·arXiv CS.AI·4h agoA
4

Distributionally Robust Listwise Preference Optimization

Research·arXiv CS.AI·4h agoA
4

SUNTA: Hierarchical Video Prediction with Surprise-based Chunking

Research·arXiv CS.AI·4h agoA
4

Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation

Hardware·arXiv CS.AI·4h agoA
4

EduArt: An educational-level benchmark for evaluating art history knowledge in large language models

Research·arXiv CS.AI·4h agoA
4

Diverse Evidence, Better Forecasts: Multi-Agent Deliberation Under Information Asymmetry

Research·arXiv CS.AI·4h agoA
4

MolSight: A Graph-Aware Vision-Language Model for Unified Chemical Image Understanding

Research·arXiv CS.AI·4h agoA
4

Understanding Geometric Representations in Self-Supervised Vision Transformers via Subspace Intervention

Models·arXiv CS.AI·4h agoA
4

Multimodal Knowledge Edit-Scoped Generalization for Online Recursive MLLM Editing

Research·arXiv CS.AI·4h agoA
4

CreativityNeuro: Steering Language Model Weights to Improve Divergent Thinking and Reduce Mode Collapse

Models·arXiv CS.AI·4h agoA
4

Towards Real-World Ultrasound Understanding: Large Vision-Language Models from Multi-Image Examinations with Long-Form Reports

Open Source·arXiv CS.AI·4h agoA
4

Model Merging as Probabilistic Inference in Fine-Tuning Parameter Space

Models·arXiv CS.AI·4h agoA
4

ReQuest: Rethinking-based Question-Aware Frame Selection for Long-Form Video QA

Models·arXiv CS.AI·4h agoA
4

QWERTY: Training-Free Motion Control via Query-Warped Video Diffusion Transformers

Research·arXiv CS.AI·4h agoA
4

SPARCLE: SPeaker-aware Aligned Representations via Contrastive Language Embeddings

Research·arXiv CS.AI·4h agoA
4

From Approximation to Emergence: A Theory of Deep Learning

Products·arXiv CS.AI·4h agoA
4

SimWorlds: A Multi-Agent System for Dynamic 3D Scene Creation

Research·arXiv CS.AI·4h agoA
4

SpaceEra++: A Unified Framework Towards 3D Spatial Reasoning in Video

Research·arXiv CS.AI·4h agoA
4

Expander Sparse Autoencoders: Parameter-Efficient Dictionaries for Mechanistic Interpretability

Research·arXiv CS.AI·4h agoA
4

Mastermind: Strategy-grounded Learning for Repository-Scale Vulnerability Reproduction

Products·arXiv CS.AI·4h agoA
4

Set Diffusion: Interpolating Token Orderings Between Autoregression and Diffusion for Fast and Flexible Decoding

Research·arXiv CS.AI·4h agoA
4

InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation

Research·arXiv CS.AI·4h agoA
4

Rethinking Speech-LLM Integration for ASR: Effective Joint Speech-Text Training by Interleaving

Research·arXiv CS.AI·4h agoA
4

Generic Expert Coverage for Pruning SparseMixture-of-Experts Language Models

Models·arXiv CS.AI·4h agoA
4

A-TMA: Decoupling State-Aware Memory Failures in Long-Term Agent Memory

Research·arXiv CS.AI·4h agoA
4

A Mathematical Introduction to Diffusion Models

Research·arXiv CS.AI·4h agoA
4

PairCoder++: Pair Programming as a Universal Paradigm for Verified Code-Driven Multimodal and Structured-Artifact Generation

Policy·arXiv CS.AI·4h agoA
4

Domain Generalization via Text-Anchored Information Bottleneck

Research·arXiv CS.AI·4h agoA
4

HaloGuard 1.0: An Open Weights Constitutional Classifier for Multilingual AI Safety

Models·arXiv CS.AI·4h agoA
4

PARTREP: Learning What to Repeat for Decoder-only LLMs

Opinion·arXiv CS.AI·4h agoA