Anthropic scales Claude; Global AI regulations emerge

Anthropic / Claude ecosystem

Anthropic Makes Claude Fable 5 a Standard Feature in Max and Team Premium Subscription Plans

Anthropic has announced that its Claude Fable 5 model is now a standard feature in its Max and Team Premium subscription plans, at 50% of weekly limits. This move ends months of temporary extensions and capacity uncertainty for users of the advanced AI model.

Frontier model providers

Alibaba Launches 2.4 Trillion-Parameter Open-Weight Qwen 3.8, Claims Near-Frontier Performance

Alibaba has announced Qwen 3.8, a 2.4 trillion-parameter open-weight model that it claims is second only to Claude Fable 5 on internal benchmarks. This marks the third major Chinese model release in three months to rival frontier Western AI systems.

AI developer tooling & infrastructure

LM Studio Launches 'Bionic,' a Coding and Document Agent for Open Models

LM Studio has released 'Bionic,' a new coding and document agent built on open-weight models. Bionic offers users control over execution environments and data retention, positioning privacy and cost as key advantages over proprietary alternatives like GitHub Copilot and Claude Code.

UserTesting Integrates Customer Research into AI Tools and Design Workflows with New MCP Server

UserTesting's July 2026 release embeds customer research directly into AI tools and design workflows via a new MCP Server and Figma plugin integration. This aims to provide faster, more trusted insights into design processes.

Cloud & platform providers

Google Launches New MCP Servers to Simplify AI Agent Integration with Cloud Services

Google has launched fully managed MCP (Managed Compute Platform) servers designed to simplify the integration of AI agents with real-world data and tools like Maps, BigQuery, Compute Engine, and Kubernetes Engine. This aims to remove friction from multi-day connector setup processes.

Amazon Web Services Launches Accelerator for Chinese 'One-Person Unicorn' Founders

Amazon Web Services (AWS) has launched a new accelerator program, 'From Idea to Frontier,' targeting 'one-person unicorns' in China. The program offers up to $100,000 in credits and mentorship, betting that AI enables solo founders to build global companies in 6-12 months.

AI policy, regulation & governance

India Developing 'Well-Calibrated' AI Law, Avoiding US and EU Models

India's Ministry of Electronics and Information Technology (MeitY) Secretary has indicated that India is developing a 'well-calibrated' AI regulatory approach. This strategy aims to avoid simply replicating the EU's prescriptive model or the US's light-touch strategy.

China Cracks Down on AI Companions, Forcing Millions to Disconnect

The Chinese government has implemented new regulations that ban AI companions for minors and restrict chatbots from encouraging emotional reliance. This affects millions of users and has led major tech platforms like ByteDance, Alibaba, and Tencent to shut down their AI companion features.

EU Adopts AI Governance Principles for Health Technology Assessments

The EU has established its first sector-specific AI governance framework for clinical evidence assessment in health technology. The new principles require transparency and human accountability to preserve scientific integrity in health technology evaluations.

EU Compels Google to Share Android and Search Data with Rival AI Assistants Under Digital Markets Act

The EU Commission has issued binding specification decisions forcing Google to open 11 Android system features to rival AI assistants and share anonymised search data with competitors. This establishes concrete compliance requirements under the Digital Markets Act.

Australian Government Tightens Rules for AI Use in Departmental Decision-Making

The Australian federal government has published a new framework requiring federal agencies to ensure AI-driven decisions are fair and accurate. This follows years of scrutiny over automated welfare payment cancellations and other algorithmic risks.

Australia Proposes 'Digital Duty of Care' Policy for Large AI Firms

The Australian Labor Government is proposing a 'Digital Duty of Care' framework that would impose new compliance obligations on large AI companies operating in Australia, including OpenAI and Anthropic. This policy aims to shift responsibility onto platforms to prevent harms.

Australian Government Announces Stricter Oversight for AI in Automated Decision-Making

The Australian government has announced a stricter oversight framework for AI in automated government decision-making. This addresses concerns over unlawful automated cancellations and algorithmic risks in social services.

Industry & market moves

Anthropic Negotiates $10 Billion AI Resource Lease with Meta

Anthropic is reportedly negotiating with Meta to lease up to $10 billion in computing resources over two years. This signals Meta's entry into the cloud computing business to diversify its revenue beyond advertising.

AI product & feature launches

OpenAI Introduces Parental Control 'Study Mode' for Teen ChatGPT Sessions

OpenAI has launched a new parent-toggleable Study Mode for ChatGPT, designed to guide teenagers through problems pedagogically rather than providing direct answers. The feature also includes expanded safety notifications for violent content.

Meta Launches AI Support Assistant and Advanced Content Enforcement Tools

Meta has rolled out a new AI support assistant capable of taking direct action on account issues and responding in under five seconds. Additionally, advanced AI content enforcement tools now operate in languages covering 98% of online users, showing improved detection of scams, impersonation, and exploitative content.

Microsoft Launches Project Perception, an AI Bug Hunter to Rival Anthropic's Mythos

Microsoft has launched Project Perception, a new AI bug hunter designed with a multi-model routing architecture. It aims to undercut Anthropic's Claude Mythos on cost and expand access to enterprises globally, bypassing Mythos' export restrictions.

Nonprofit Current AI Launches Open-Source Multilingual Chatbot 'Alpha Chat'

Current AI, a nonprofit supported by $400 million in public and philanthropic funding, has launched Alpha Chat. This open-source multilingual AI chatbot aims to build public-interest AI infrastructure accessible to underserved global communities.

Nubia Announces NaviX Ultra, the World's First AI Agent Smartphone with On-Device Processing

Nubia has announced the NaviX Ultra, claiming it to be the world's first AI-agent smartphone with on-device AI processing. The device runs the Doubao mobile assistant, designed to perform multi-step tasks across apps without relying on cloud processing.

Research with immediate practical relevance

Google DeepMind Proposes Video Generators as 'World Models' for Computer Vision

Google DeepMind argues that existing video generators already contain 'world models' that could significantly advance computer vision. Their GenCeption model demonstrates matching specialized models on depth estimation and segmentation with 7-500x less training data.

Open-Weight AI Models Rapidly Approaching Frontier Cyber Skill, AISI Finds

A new report from the AI Security Institute (AISI) reveals that open-weight AI models now trail frontier closed systems by only 4–7 months on offensive cybersecurity tasks, a reduction from 6–10 months in 2025. Autonomous cyberattacks using Chinese open models reportedly cost under $2 on simulated enterprise networks.

VIDRAFT Releases Aether-7B-5Attn, First Fully Open Korean Language Foundation Model

VIDRAFT (FINAL-Bench / SeaWolf-AI) has released Aether-7B-5Attn, a 6.59 billion-parameter, fully open-source Korean language foundation model. It includes all training data, code, and checkpoints, making it the first built by a single AI startup in Korea.

Security Researcher Demonstrates Easy Open-Weight AI Model Poisoning for Under $100

A cybersecurity researcher has demonstrated that open-weight AI models can be poisoned with just 10 training examples and under $100 in compute, with minimal detection. This raises significant trust and supply-chain security concerns for fine-tuned models.

VideoChat3 Outperforms GPT-5 on Video Grounding, Open-Source Training Stack Released

A 4-billion-parameter open-source video understanding model, VideoChat3, has been released with its complete training code and datasets. Developed by a consortium of Chinese universities, it outperforms GPT-5 and Gemini 2.5 Flash on temporal grounding benchmarks while using half the visual tokens.