Anthropic Takes Cutting-Edge "Mythos" Model Offline — SK Telecom Access Revocation Order Revealed

Just before Anthropic took its cutting-edge AI model "Claude Mythos" offline, it emerged that the White House had ordered the company to revoke South Korea's SK Telecom's access. The order was reportedly motivated by concerns over SK Telecom's ties to China.

The report came exclusively from WIRED and stands as a significant case study in how U.S. AI export controls are being applied at the level of specific companies. Some have suggested that this government pressure may have played a role in Anthropic's decision to take its frontier model entirely offline.

Claude Design Gets Major Upgrades — Two-Way Integration with Code, Canva/Adobe Export Support

Anthropic has significantly expanded the beta capabilities of its design tool, "Claude Design." The key improvements include:

The ability to import multiple design systems and maintain design consistency across projects

Seamless two-way integration with "Claude Code"

Greatly expanded export connectors for external tools such as Adobe and Canva

The update strengthens the core design premise: letting you complete the entire workflow—from design to code to export into external tools—entirely within the Claude ecosystem.

China vs. G7 — Geopolitical Clash Over Providing AI to Developing Countries

While China offers free AI technology to developing countries, G7 nations continue negotiating over access rights to American-made AI models. According to The Next Web, this confrontation has become a race to expand influence under the banner of "democratizing AI."

China champions open cooperation, while the G7 faces the difficult task of deciding who gets access to which models within the framework of export controls. The situation underscores how AI technology has become a new instrument of geopolitical influence.

AI Medical Tool Matches Physicians — Evidence Reported by the Financial Times

According to the Financial Times, AI-based medical advice tools are achieving accuracy on par with—or even exceeding—physicians' advice. For details on the specific evaluation methodology, you'll need to refer to the original article, but it confirms that AI's practical utility in medicine continues to be validated.

Qwen Lead Lin Junyang Founds New AI Lab, Raises Funding at $2 Billion Valuation

Lin Junyang, who led development of the open-source LLM "Qwen" series, has founded a new AI lab and completed a funding round at a $2 billion valuation. Within the LocalLLaMA community, the news is widely seen as a positive development for open-source/open-weight models. All eyes are on what this leader—the force behind Qwen—will build next.

Ultra-Compact TTS Model "Inflect-Nano" Released — Just 4.63M Parameters

A developer known as b111ue has released "Inflect-Nano-v1," a text-to-speech (TTS) model with only 4.63M parameters. At roughly 1/17 the size of Kokoro and about 1/1000 the size of Fish Audio S2 Pro, it can reportedly generate 24kHz audio and run locally even on very low-spec environments.

While it has some quality shortcomings, it's an intriguing result as a baseline for speech synthesis on edge devices and in browser/WASM environments. The model is available on Hugging Face.

The "Learning Penalty" of Generative AI — Evidence from Chinese Secondary Schools

A paper published by CEPR (Centre for Economic Policy Research) reports evidence that generative AI use in China's secondary education produces a "learning penalty." The study examined what differences in creativity and learning outcomes emerge between students who use AI and those who don't, providing an important data point in the debate over how AI should be used in education.

---