Overview
- Huang said on a podcast that China has idle, fully powered data centers and ample chips that could be pooled to train systems on par with Anthropic’s Mythos.
- He pointed to China producing about 60% of mainstream chips and hosting roughly half of AI researchers, adding that many less advanced chips networked together can still deliver strong AI training performance.
- Anthropic’s internal tests found Mythos could surface thousands of software flaws in browsers and operating systems, and the AI Security Institute said the model could run autonomous multi‑stage network attacks, prompting Anthropic to restrict access.
- Reuters has reported that banks relying on decades‑old software face higher exposure to AI‑assisted breaches, and Treasury Secretary Scott Bessent recently met major bank executives after calling the risks unprecedented.
- Huang urged dialogue over blanket restrictions to set guardrails on such tools, a stance voiced after Anthropic reported last November that Chinese state actors tried to misuse its Claude Code assistant.