AI news on Friday, June 26, 2026

SIGNIFICANCE

★ Top story · Models2d ago

Anthropic’s Mythos 5 AI Model Cleared by US for Wider Use

The U.S. Department of Commerce has authorized the use of Anthropic’s Mythos 5 artificial intelligence model for a select group of trusted partners. By easing previous restrictions on this high-capability model, regulators are allowing deeper collaboration on sensitive projects while maintaining oversight. This decision marks a shift in how the government manages the deployment of advanced AI, balancing the need for safety with the desire to foster innovation within controlled environments.

BBloomberg Technology BBloomberg Technology TThe Decoder Read original ↗

Friday — June 26, 2026

Anthropic’s Mythos 5 AI Model Cleared by US for Wider Use

Previewing GPT-5.6 Sol: a next-generation model

Apple’s Vision Pro and Smart Glasses Chief to Join OpenAI

Ford rehires ‘gray beard’ engineers after AI falls short

What Is AI Distillation and Why Is It a Worry for the Industry?

Linux Foundation and 20 tech giants launch Akrites to fix open-source flaws before AI-powered attacks hit

Anthropic Economic Index report: Cadences

Anthropic doesn't need junior engineers anymore thanks to AI and warns of an economic shock when other industries follow

OpenAI Hires Uber’s Prabhjeet Singh to Lead India Operations

Anthropic Moves Toward Deal With US to Lift Curbs on AI Models

AI startup Lindy ditched Claude entirely for Deepseek, saving millions as cost pressure mounts on Anthropic

ConflictScore: Identifying and Measuring How Language Models Handle Conflicting Evidence

SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference

Where Larger Models Excel: The Primacy of Constraint-Guided Reasoning

Deterministic Pareto-Optimal Policy Synthesis for Multi-Objective Reinforcement Learning

Instruction Bleed: Cross-Module Interference in Prompt-Composed Agentic Systems

Staying VIGILant: Mitigating Visual Laziness via Counterfactual Visual Alignment in MLLMs

A Generalization Theory for JEPA-Based World Models

At the Edge of Understanding: Sparse Autoencoders Trace The Limits of Transformer Generalization

Modeling Local, Global, and Cross-Modal Context in Multimodal 3D MRI

Bridging Vision and Language Concepts through Optimal Transport Semantic Flow

Learning to Recover Task Experts from a Multi-Task Merged Model

Optimizing CUDA like a Human: Micro-Profiling Tools as Expert Surrogates for LLM-Based GPU Kernel Optimization

The Capability Frontier: Benchmarks Miss 82% of Model Performance

Perception, Verdict, and Evolution: Hindsight-Driven Self-Refining Forensics Agent for AI-Generated Image Detection

Erase-then-Delta Attention: Decoupling Erase and Write Addresses in Delta-Rule Linear Attention

Unison: Benchmarking Unified Multimodal Models via Synergistic Understanding and Generation

Closing the Quality Gap in Low-Resource Text-to-Speech: LoRA Fine-Tuning of VoxCPM2 for Khmer and Korean

Reasoning Quality Emerges Early: Data Curation for Reasoning Models

Radical AI Interpretability

Robust Onion: Peeling Open Vocab Object Detectors Under Noise

EVOM: Agentic Meta-Evolution of Actor-Critic Architectures for Reinforcement Learning

DocArena: Turning Raw Documents into Controllable Training Environments for Document Search Agents

Quantization in Federated Learning: Methods, Challenges and Future Directions

Detecting and Controlling Sycophancy with Cascading Linear Features

EGG: An Expert-Guided Agent Framework for Kernel Generation

Normalizing Flows are Capable Models for Continuous Control

auto-psych: Automating the science of mind using agent-driven theory discovery and experimentation

Mask to Concept: Auto-Promptable SAM3 via Efficient Test-Time Concept Embedding Search for Few-Shot Annotation

AlgoEvolve: LLM-driven Meta-evolution of Algorithmic Trading Programs

MKG-RAG-Bench: Benchmarking Retrieval in Multimodal Knowledge Graph-Augmented Generation

Full spectrum Unlearnable Examples via Spectral Equalization

Rethinking Training & Inference for Forecasting: Linking Winner-Take-All back to GMMs

AIGP: An LLM-Based Framework for Long-Term Value Alignment in E-Commerce Pricing

Refusal Lives Downstream of Persona in Chat Models

Thinking Like a Scientist? A Structural Study of LLM-Generated Research Methods

OpenFinGym: A Verifiable Multi-Task Gym Environment for Evaluating Quant Agents

Do Image Editing Models Understand Lighting?

PhysRAG: Enhancing Physics-Awareness in Video Generation via Retrieval-Augmented Generation

ViQ: Text-Aligned Visual Quantized Representations at Any Resolution

Safe Autoregressive Image Generation with Iterative Self-Improving Codebooks

Where Do Models Find Happiness? Emotion Vectors in Open-Source LLMs

Confidence-Aware Tool Orchestration for Robust Video Understanding

Nemotron-TwoTower: Diffusion Language Modeling with Pretrained Autoregressive Context

Language-Based Digital Twins for Elderly Cognitive Assistance

A Causal Foundation Model for Structure and Outcome Prediction

CARVE: Content-Aware Recurrent with Value Efficiency for Chunk-Parallel Linear Attention

Discovering Millions of Interpretable Features with Sparse Autoencoders

OpenRCA 2.0: From Outcome Labels to Causal Process Supervision

OPID: On-Policy Skill Distillation for Agentic Reinforcement Learning

HyperDFlash: MHC-Aligned Block Speculative Decoding with Gated Residual Reduction

SamaVaani: Auditing and Debiasing Multilingual Clinical ASR for Indian Languages

Life After Benchmark Saturation: A Case Study of CORE-Bench

ReaORE: Reasoning-Guided Progressive Open Relation Extraction Empowered by Large Reasoning Models

SemiAnalysis Xie on Asia AI Supply Chain

State Street: Asia May Win in AI Boom

LiveEdit: Towards Real-Time Diffusion-Based Streaming Video Editing

RedVox: Safety and Fairness Gaps in Speech Models Across Languages

SKILL-DISCO: Distilling and Compiling Agent Traces into Reusable Procedural Skills

HauntAttack: When Attack Follows Reasoning as a Shadow

CSU-PCAST: A Dual-Branch Transformer Framework for medium-range ensemble Precipitation Forecasting

Epiphany-Aware KV Cache Eviction Without the Attention Matrix

KARLA: Knowledge-base Augmented Retrieval for Language Models

Fast LeWorldModel

Run a vLLM Server on HF Jobs in One Command

GEOALIGN: Geometric Rollout Curation for Robust LLM Reinforcement Learning

Qwen-Image-Agent: Bridging the Context Gap in Real-World Image Generation

AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems

DiCoBench: Benchmarking Multi-Image Fine-Grained Perception via Differential and Commonality Visual Cues