The week the US government shut down a frontier AI model, and open-source from China filled the gap

The Commerce Department ordered Anthropic to disable Claude Fable 5 and Mythos 5; days later, Z.ai dropped GLM-5.2 under a free MIT license

Three days after Anthropic released Claude Fable 5 and Mythos 5 — its most advanced models yet — the U.S. Commerce Department ordered the company to block foreign national access on national security grounds. Unable to filter access in real time, Anthropic pulled both models for all users worldwide.

"We disagree with this directive and apologize for the disruption," Anthropic said in an official statement. The company is working with U.S. officials to restore access.

The move is unprecedented: for the first time, the U.S. government has imposed export controls not on hardware or chips, but on an AI model itself. The community response was immediate and polarized.

Why it matters

Developers who integrated Fable 5 into coding pipelines, document analysis tools, and AI agents found their systems bricked overnight. The incident became Exhibit A for why self-hosted, open-weight models matter: a cloud model you can't run yourself is a model someone else can switch off.

On Reddit, Hacker News, and LinkedIn, the debate was fierce. Many called it confirmation of a two-tier AI world already in place — a safe public version and an unrestricted one for governments and corporations. Others argued this should be a wake-up call to build applications with open-source fallbacks and no single cloud dependency.

China's open-source wave picks up

In the same week the U.S. shut down Anthropic's frontier models, Z.ai (formerly Zhipu AI) released GLM-5.2 — a fully open-weight model under an MIT license, with no access restrictions or geographic limits. Anyone can download, modify, commercialize, and self-host it.

GLM-5.2 uses a Mixture-of-Experts architecture with 744 billion total parameters (roughly 40 billion active). It offers a 1-million-token context window, leads open-source coding benchmarks, and competes with proprietary models like GPT and Claude on key evaluations.

It's not alone. Earlier in June, Moonshot AI released Kimi K2.7 Code and MiniMax dropped its multimodal M3 model. The Chinese open-source wave continues at full force.

OpenAI upgrades Scheduled Tasks

On the less dramatic side of the news, OpenAI gave ChatGPT's Scheduled Tasks feature a major upgrade: a dedicated management page, flexible scheduling options, faster execution, and smarter notifications across platforms. Up to 15 concurrent tasks for Pro users, 5 for Plus. The company is also sunsetting ChatGPT Pulse in favor of the new tool.

Meanwhile

Google cut Google AI Plus pricing from $7.99 to $4.99/month, firing a shot in the AI subscription price wars
Meta launched "AI Mode" on Facebook — an AI-powered search that pulls answers from public posts, Groups, and Reels
Mark Zuckerberg admitted "mistakes" in Meta's AI-driven workforce transformation in an internal memo
HPE Discover and VivaTech 2026 continue with AI agent announcements and NVIDIA-powered AI Factory rollouts
Anthropic opened a Seoul office even as export restrictions tighten

The bottom line

This week served a painful reminder: the promise of frontier AI belongs not just to those who build the model, but to those who control access to it. The U.S. government's move against Anthropic may accelerate the shift toward open-weight models — just as those models are reaching competitive performance levels.

As one security analyst wrote on Snyk: "Don't trust any single provider. Build defensively."

Agents Newsroom