AI news on Thursday, July 2, 2026

SIGNIFICANCE

★ Top story · Policy2h ago

OpenAI Proposes Giving the US Government a 5% Stake, FT Says

OpenAI has reportedly entered preliminary discussions regarding granting the United States government a 5% equity stake in the company. This proposal marks a notable shift in the firm's engagement with federal oversight as it seeks to address national security concerns related to advanced artificial intelligence development.

BBloomberg Technology TThe Decoder BBloomberg Technology Open full story →Read original ↗

Thursday — July 2, 2026

OpenAI Proposes Giving the US Government a 5% Stake, FT Says

StateFlow: Dual-State Recurrent Modeling for Long-Horizon Time Series Forecasting

Apple Extends Private Cloud Compute to Google Cloud for the First Time

A Task-State Representation for Long-Horizon Mobile GUI Agents

MSQA: A Natively Sourced Multilingual and Multicultural SimpleQA Benchmark

Readable but Not Controllable: Neuron-Level Evidence for Medical LLM Hallucination

Identifying and Resolving Pitfalls of Knowledge-Based VQA Benchmarks: Auditing, Repairing, and Augmenting

MosaicKV: Serving Long-Context LLM with Dynamic Two-D KV Cache Compression

Self-conditioned Flow Map Language Models via Fixed-point Flows

Self-GC: Self-Governing Context for Long-Horizon LLM Agents

GEAR-Seg: A Grounded Explainable Agent for Reasoning Segmentation and Data Engine

MindEdit-Bench: Benchmarking Object-Level Counterfactual Spatial Reasoning in VLMs from In-the-Wild Photos

Low Perplexity is Repetition: A One-Dimensional Self-Conditioning Attractor in Continuous Diffusion LMs

LLM-Guided ODE Discovery and Parameter Inference from Small-Cohort Aggregate Data

PHREEQC-MCQ-200: A Diagnostic Benchmark for Tool-Augmented Scientific Simulator Agents

Synergistic Perception-Reasoning Governance: Grounding Medical MLLMs with Verifiable Anatomical Evidence

Recovering Input Text from Hidden States: Study of Gradient-Based Inversion of Decoder-Only Language Models

Representation as a Bottleneck for Mechanistic Interpretability: The Manifestation Unit Protocol

Multi-scale Mixture of World Models for Embodied Agents in Evolving Environments

EFlow: Learning Evidence Flow for Long-Video Reasoning with Adaptive Reflection

Understanding Why Language Models Hallucinate: Testing Reasoning Against Priors

Active Spatial Guidance: Eliminating Injected Positional Mechanisms in Vision Transformers

AdaBoosting Text Prompts for Vision-Language Models

Learning Generalizable Skill Policy with Data-Efficient Unsupervised RL

SLIM-RL: Risk-Budgeted Random-Masking RL for Diffusion LLMs Without Trajectory Slicing

Explainable AI for Cancer Drug Response Prediction: Beyond Univariate Feature Attributions

Multi-Hypothesis Test-Time Adaptation to Mitigate Underspecification

DriveVer: Lightweight Trajectory Evaluator as Test-Time Verifier for Autonomous Driving

The Illusion of High Utility in Safety Alignment of Text-to-Image Diffusion Models

LIST3R: Long-sequence Instance-aware 3D Reconstruction

Not All Prediction Targets Keep Training-Free Diffusion Guidance on the Manifold

Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity

ABot-M0.5: Unified Mobility-and-Manipulation World Action Model

EPC: A Standardized Protocol for Measuring Evaluator Preference Dynamics in LLM Agent Systems

Meta Is Charging a Subscription for Smart Glasses Features. Welcome to the New Era of Consumer Tech

Vitality-Aware Compression for Efficient Image-to-Shape Diffusion Transformers

Self-Evolving Agents with Anytime-Valid Certificates

AEGIS: A Multi-Task Joint-Embedding Predictive Architecture for Mammography

Multimodal Continuous Reasoning via Asymmetric Mutual Variational Learning

Creating Impactful Autonomous Driving Datasets: A Strategic Guide from Research Gap to Benchmark

Post-Training Pruning for Diffusion Transformers

Validating Causal Abstraction Metrics on Simulated Complex Systems

LeVLJEPA: End-to-End Vision-Language Pretraining Without Negatives

Graph-Native Reinforcement Learning Enables Traceable Scientific Hypothesis Generation through Conceptual Recombination

Auditing Forgetting in Limited Memory Language Models

Making Failure Safe: A Constrained, Verifiable Agent Framework for Open-Web Data Collection

Pano2World: End-to-End 3D Generation via Unified Multi-View Sequences

VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement

PixelEyes: Decoupling Perception and Reasoning for Pinpoint Visual Evidence Seeking

Generative Modeling of Quantum Distribution with Functional Flow Matching

Coachable agents for interactive gameplay

EgoSafetyBench: A Diagnostic Egocentric Video Benchmark for Evaluating Embodied VLMs as Runtime Safety Guards

Training-Free Debiasing of Diffusion Models via CLIP-Guided Denoising Optimization

Information-Regularized Attention for Visual-Centric Reasoning

SAP Restricts Hiring, Travel to Fund ‘Significant’ AI Push

Rosetta: Composable Native Multimodal Pretraining

NeuroCogMap Reveals Cognitive Organization of Large Language Models

SoftBank Plans AI Cloud Services in US to Tap Surging Demand

A Mechanism-Driven Theory of Phase Transitions in Active Learning

BaseRT: Best-in-Class LLM Inference on Apple Silicon via Native Metal

Multi-Turn Agentic Scientific Literature Search via Workflow Induction

Testing Frontier Large Language Models' Physics Literacy in Parallel Physical Worlds

Soft Mixture-of-Recursions: Going Deeper with Recursive Vision Transformers

A Contextual-Bandit Oversight Game with Two-Sided Informational Asymmetry

Beyond Document Grounding: Span-Level Hallucination Detection over Code, Tool Output, and Documents

FRAME: Learning the Adaptation Domain with a Mixture of Fractional-Fourier Experts

Beyond Activation Alignment:The Alignment-Diversity Tradeoff in Task-Aware LLM Quantization

Loss Smoothing for Stable Adaptation Under Distribution Shift

Dynamic Bidirectional Pattern Memory: A Production-Scale Empirical Characterisation of Inference-Time Gating in Clinical NLP

CAT: Confidence-Adaptive Thinking for Efficient Reasoning of Large Reasoning Models

Rethinking Multi-Label Image Classification With Deep Learning: Taxonomy, Challenge, and Outlook

TrajLoc: Trajectory-Attention Localization for Multi-Object Motion Control

MoVA: Learning Asymmetric Dual Projections for Modular Long Video-Text Alignment

AI Native Games: A Survey and Roadmap

Two AI Metrics Diverged: Will it Make All the Difference?

SpiralFovea: Input-Adaptive Foveated Tokenization as a Third Lever of Resource-Adaptive Inference

GaussianFusion: Unified 3D Gaussian Representation for Multi-Modal Fusion Perception

From Pixels to Temporal Correlations: Learning Informative Representations for Reinforcement Learning Pre-training

GKDT: General Keypoint Detection Transformer