Global Anthropic Ban Prompts Launch of Fugu and Tulongfeng
Sakana AI and 360 Security released specialized models to fill the global supply gap caused by strict U.S. export controls on Anthropic's Claude Mythos 5.
On June 27, 2026, Sakana AI and 360 Security launched new AI systems to replace Anthropic’s flagship models, following a two-week U.S. export ban on Claude Fable 5 and Mythos 5. The emergency directive barred foreign nationals from accessing Anthropic’s newest models due to national security risks. This hard cutoff forced developers outside the United States to immediately seek local alternatives for frontier workloads.
The Anthropic Supply Gap
On June 12, the Trump administration ordered Anthropic to pull Fable 5 and Mythos 5 offline globally because the company could not verify user nationalities in real-time. The U.S. government eased the restriction on June 26, allowing approximately 100 trusted U.S. organizations managing critical infrastructure to restore access to Mythos 5. For the rest of the world, the export directive remains in full effect.
Compliance concerns have already shifted corporate behavior. Major financial institutions in Hong Kong, including JPMorgan Chase and Goldman Sachs, disabled access to all frontier U.S. models to avoid violating strict interpretations of the export rules.
Before the ban, the Mythos-class models established a distinct performance tier above Claude Opus. The high pricing reflected this capability jump, with Mythos 5 costing $10 per million input tokens and $50 per million output tokens.
| Benchmark | Mythos 5 / Fable 5 | Opus 4.8 |
|---|---|---|
| SWE-Bench Pro | 80.3% | 69.2% |
| FrontierCode (Diamond Split) | 29.3% | ~14.6% |
| CursorBench 3.1 | 72.9% | Not specified |
Autonomous offensive cyber capabilities, including the ability to discover zero-day vulnerabilities in browsers and operating systems, directly triggered the federal intervention.
Asian Model Launches
Tokyo-based Sakana AI released Fugu, an orchestration model named after the Japanese blowfish and built on a 7-billion-parameter architecture. Instead of processing complex tasks inside a single massive neural network, Fugu uses a unified API to coordinate multiple external models and AI agents. Sakana claims this architecture matches Fable 5 on key benchmarks, marketing the system explicitly as a way to access frontier capabilities without export risk. For teams evaluating multi-agent coordination, Fugu offers a structural alternative to monolithic models.
Beijing-based 360 Security introduced Tulongfeng, a vulnerability discovery tool positioned as a direct replacement for the cybersecurity-focused Mythos series. Alongside it, the company launched Yitianzhen, an automated cyber defense system. These specialized tools target the exact autonomous security capabilities that the U.S. government sought to contain.
These late-June releases follow Zhipu AI’s launch of GLM-5.2, which debuted just one day after the initial Anthropic directive. Developers have increasingly adopted GLM-5.2 as an open-weight coding assistant due to its top-three global benchmark ranking and low operational costs. Industry analysts have hailed the release as a “DeepSeek moment” for the Chinese AI sector, proving that local labs can rapidly deploy cost-effective models that rival Western flagships.
If you run production workloads on frontier U.S. models, geopolitical availability is now a hard infrastructure constraint. You must design your routing layers to fail over to local or open-weight models when API access is suddenly revoked.
Get Insanely Good at AI
The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.
Keep Reading
How to Distill Large Models With CompactifAI Top-K Logits
Learn how to use Multiverse Computing's CompactifAI to distill large language models locally using offline Top-K logits and Fused Chunked KL loss.
Chinese AI labs are trimming offensive cyber skills from new models as Beijing drafts export curbs
Bloomberg reported on November 16 that Alibaba, Tencent and Moonshot have shipped recent models with reduced offensive cyber capability after a Ministry of Commerce draft proposed banning exports of AI attack tools. The methods differ, and only Moonshot published a number.
US closes the cloud loophole: new BIS rule licenses AI compute for China-linked customers and five diversion hubs
BIS's November 16 interim final rule requires a license for advanced AI chips going to any data center that will give China- or Macau-headquartered firms more than 12 million aggregate TPP, adds license requirements for Brazil, Indonesia, Malaysia, Singapore and Thailand, and formally buries the AI Diffusion Rule.
Blackwell fingerprints in DeepSeek, Kimi and MiniMax weights: how checkpoint forensics reads the GPU behind a model
A November 12 report by former OpenAI researcher Will DePue says six open-weight models from four Chinese labs were initialized on export-controlled Nvidia Blackwell GPUs, using leftover random-number patterns in the published weights. Here is how the method works, what it proves and what it does not.
Jensen Huang resigns as Nvidia CEO: Colette Kress takes over, and the AI industry loses its backstop of last resort
Nvidia said on November 12 that Huang had resigned as CEO and director, effective immediately, with CFO Colette Kress as interim CEO. The exit puts more than $100 billion in lease guarantees, a $12.9 billion Hugging Face deal and US chip policy in new hands days before earnings.