AI News — Friday, June 26, 2026

10
White House Asks OpenAI to Pause New Model Release Over Safety Concerns

The White House has reportedly requested OpenAI to delay the release of its upcoming model, citing significant safety concerns.

TechCrunchindustry
9
Patronus AI Secures $50M to Develop 'Digital Worlds' for Stress-Testing AI Agents

Patronus AI has raised $50 million to create virtual environments designed to rigorously test and validate the safety and reliability of AI agents.

TechCrunchindustry
9
Anthropic's Claude Gains Ground on ChatGPT, Attracting Paid Consumers

Anthropic's Claude is increasingly capturing the paid consumer market, challenging ChatGPT's dominance in the competitive landscape of large language models.

TechCrunchindustry
9
Google Research's AMIE Shows Promise in Managing Health Conditions

New research from Google demonstrates how AMIE, their medical AI, could effectively assist in the management of various health conditions.

Google AI Blogresearch
8
Are We Ready For An Agent-Native Memory System?

This paper explores the foundational requirements and challenges for developing memory systems specifically designed for AI agents.

Hugging Faceresearch
8
Nous Research Releases NousCoder-14B, an Open-Source Coding Model

Nous Research has launched NousCoder-14B, an open-source coding model positioned to compete with proprietary solutions like Claude Code.

VentureBeatopen-source
8
OpenAI Details How AI Agents Are Transforming Workflows

OpenAI shares insights into the evolving role of AI agents and their significant impact on transforming various aspects of work.

OpenAI Blogindustry
7
DomainShuttle: Advancing Freeform Open-Domain Subject-Driven Text-to-Video Generation

Researchers introduce DomainShuttle, a new method for generating high-quality, subject-driven videos from text prompts across diverse domains.

Hugging Faceresearch
7
Wan-Streamer v0.1: Real-time Interactive Foundation Models for End-to-End Applications

Wan-Streamer v0.1 is presented as a new framework enabling end-to-end real-time interaction with foundation models.

Hugging Faceproduct
7
Unlimited-OCR: One-shot Long-horizon Optical Character Recognition Released

Baidu has open-sourced Unlimited-OCR, a novel system designed for one-shot, long-horizon optical character recognition tasks.

Lobste.rsopen-source
6
ShutterMuse: Capture-Time Photography Guidance with Multimodal Large Language Models

ShutterMuse introduces a system that leverages Multimodal Large Language Models to provide real-time guidance for photography during the capture process.

Hugging Faceresearch
6
Improved Large Language Diffusion Models for Enhanced Generation

This paper details advancements in large language diffusion models, leading to improved performance in generative tasks.

Hugging Faceresearch
6
MVTrack4Gen: Multi-View Point Tracking as Geometric Supervision for 4D Video Generation

MVTrack4Gen proposes using multi-view point tracking as a geometric supervision method to improve the generation of 4D videos.

Hugging Faceresearch
5
The Thinking Engineer Toolkit Goes Live, Thanks to DEV Community Support

The 'Thinking Engineer Toolkit,' a new resource for developers, has been launched following positive engagement and feedback from the DEV community.

Dev.toproduct
5
Causal-rCM: A Unified Recipe for Autoregressive Diffusion Distillation in Streaming Video Generation

Causal-rCM presents a unified approach for autoregressive diffusion distillation, applicable to streaming video generation and interactive world models.

Hugging Faceresearch