AI News — Tuesday, July 21, 2026

10
Anthropic's Landmark $1.5B Copyright Settlement Approved

A significant $1.5 billion copyright settlement involving AI company Anthropic has been officially approved, setting a major precedent for intellectual property in the AI industry.

TechCrunchindustry
9
Google Developing New AI Chip to Enhance Gemini Efficiency

Google is reportedly working on a new proprietary AI chip specifically designed to improve the efficiency and performance of its Gemini AI models.

TechCrunchindustry
8
Trump's Latest AI Czar Resigns Shortly After Appointment

The recently appointed AI czar for the Trump administration has already resigned, indicating potential instability or challenges in high-level AI policy leadership.

TechCrunchindustry
8
OpenAI Addresses Safety and Alignment for Long-Horizon AI Models

OpenAI has published insights into its ongoing efforts to ensure safety and alignment for advanced, long-horizon AI models, highlighting critical challenges in future AI development.

OpenAI Blogresearch
7
RESOURCE2SKILL: Distilling Executable Agent Skills from Multimodal Resources

A new research paper introduces RESOURCE2SKILL, a method for distilling executable agent skills directly from human-created multimodal resources, promising more capable AI agents.

Hugging Faceresearch
7
RAGU: A Multi-Step GraphRAG Engine with Compact Domain-Adapted LLM

RAGU presents a novel multi-step GraphRAG engine that incorporates a compact, domain-adapted LLM for enhanced retrieval-augmented generation, improving information retrieval accuracy.

Hugging Faceresearch
7
Xiaomi-Robotics-1: Scaling Vision-Language-Action Models with 100K Hours of Real-World Data

Xiaomi-Robotics-1 demonstrates a significant advancement in vision-language-action models by leveraging over 100,000 hours of real-world robotic trajectories for improved performance.

Hugging Faceresearch
7
AI and Code Ownership: Who is Responsible for Generated Code?

This article delves into the complex legal and ethical questions surrounding code ownership and responsibility when AI systems generate software, a growing concern for developers and companies.

Dev.toindustry
6
Cura 1T: Specialized Model for Agentic Healthcare

Cura 1T introduces a specialized 1-trillion parameter model designed for agentic applications within the healthcare sector, aiming to enhance automated medical tasks.

Hugging Faceresearch
6
Google Expands App Connectivity to Search for Enhanced User Experience

Google is enabling more applications to connect directly with its Search platform, allowing users to find and interact with app content more seamlessly through search queries.

Google AI Blogproduct
6
ReflectionCLI 2.0: Local-First Thinking CLI for AI-Assisted Development

ReflectionCLI 2.0 is released as a local-first command-line interface designed to facilitate AI-assisted development with enhanced thinking and reflection capabilities.

Dev.toopen-source
5
The Smolagents Bug: A Case Study in Agent Retrying Valid Code

A developer recounts a frustrating bug in the smolagents framework where an AI agent repeatedly retried the same valid code, highlighting challenges in agent reliability and debugging.

Dev.toopen-source
5
Triton Language for Alibaba SAIL Announced

Alibaba's T-Head group has announced support for the Triton language within its SAIL platform, potentially boosting high-performance AI kernel development for their hardware.

Lobste.rsopen-source
5
NYC AI Summit: Educators and Industry Leaders Shape AI's Future in Classrooms

New York City hosted a summit where educators and industry leaders collaborated at Google's offices to discuss and shape the integration of AI into future classroom environments.

Google AI Blogindustry
5
S1-Omni: A Unified Multimodal Reasoning Model for Scientific Understanding

S1-Omni introduces a unified multimodal reasoning model designed to enhance scientific understanding, prediction, and generation across various data types.

Hugging Faceresearch