Anthropic's Fable 5 Export Restrictions Fully Lifted — Available Worldwide Starting July 1

On June 12, the U.S. government imposed export restrictions on Anthropic's latest models, Claude Fable 5 and Claude Mythos 5. With no way to verify users' nationalities in real time, Anthropic had no choice but to suspend access to both models for all users.

On June 30, however, these export restrictions were **fully lifted**. According to Anthropic's official announcement, Fable 5 will be available to users worldwide starting Wednesday, July 1 on Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Pro, Max, Team, and some Enterprise plans will have access at up to 50% of weekly usage limits through July 7, after which availability will switch to usage-based credits.

As for Mythos 5, access was restored for some U.S. organizations following U.S. government approval on June 26, and Anthropic is reportedly now coordinating an expansion of access to a broader range of domestic and international partners through the Glasswing program.

The regulatory turmoil caused roughly three weeks of disruption across the industry, but this brings the matter to a close for now.

Vinton Cerf, Father of the Internet, to Retire from Google

Vinton Cerf, known as the co-designer of the TCP/IP protocol and often called the father of the internet, will step down as Google's Chief Internet Evangelist next week.

Since joining Google in 2005, Cerf has been a driving force behind the internet's adoption and growth. According to TechCrunch, his retirement is drawing attention as an event that marks the end of an era. At a moment when the center of gravity in technology is shifting from the internet to AI, the departure of a figure who represents the field's very origins carries considerable symbolic weight.

Vertiv Opens Its First Southeast Asia Plant for AI Data Centers in Malaysia

Vertiv Holdings, a major U.S. provider of data center equipment, has opened a new plant in Johor, Malaysia. According to Bloomberg, the facility sits just north of Singapore and will manufacture power and cooling systems for customers in Southeast Asia, North Asia, Australia, and New Zealand. Full operation is scheduled for 2027.

This marks Vertiv's first plant in Southeast Asia. With demand for AI data centers exploding, infrastructure buildout across the Asia-Pacific region is accelerating. Johor, which sits right next to Singapore, is rapidly emerging as a data center hub.

A Warning from Europe — The U.S. Can Shut Down the World's AI

A notable column published on Euronews points to the fragility of Europe's AI sovereignty. The message — that the United States can switch AI off, and that Europe must hurry to change course — is grounded in the risk that the Fable 5 export restriction episode made startlingly real.

Indeed, the episode in which Anthropic cut off access for all users at the behest of the U.S. government underscored just how easily AI technology can be disrupted by the policy decisions of a single country. The importance of building Europe's own AI capabilities is being urged more strongly than ever.

Qwen-AgentWorld — A World Model That Makes Agents Stronger Without Training Them

Qwen-AgentWorld, released by Alibaba's Qwen team in late June, is a fascinating approach that improved agent performance across seven benchmarks despite doing almost no training on agent action outputs.

The standard way to strengthen an AI agent is to train the agent itself; Qwen-AgentWorld instead trains a world model on the agent's behalf, enabling the agent to make better decisions. It's a reversal of the usual thinking — deepening the environment's understanding rather than the agent's behavioral patterns — and it could point agent development in a new direction going forward.

Sibyl — A Self-Hosted Memory System for AI Coding Agents

For developers running many AI coding agents in parallel, there's a growing need for infrastructure that lets agents share memory across the board. Sibyl, now released as open source, is a self-hostable cross-agent memory system built on SurrealDB.

It integrates with agents via CLI or MCP and comes with a shared task board, a crawler, and RAG capabilities. With multiple agents able to reference the same context, parallel work stays consistent, and it should also help prevent the generation of so-called slop.

---