Despite Warnings Global AI Development Is Racing On

AI pioneers, safety researchers, and even AI CEOs themselves have spoken of the need to slow down rapid AI development to ensure safety and survival of the human race. Despite the rhetoric, there seems to be no signs of slowing down from any front. US President Donald Trump almost in a diktat has told AI leaders that there is no way the US frontier labs can afford to slow down because "whoever wins AI, wins". Trump's main concern is China racing ahead.

And just like the US frontier labs including Anthropic, OpenAI, Meta, SpaceXAI and Google DeepMind that have been releasing new and more capable models despite, China too is racing on.

According to a Nikkei tabulation, 10 of China's leading AI companies - including DeepSeek, Alibaba/Qwen, Moonshot AI, Zhipu/GLM, Tencent, Xiaomi - launched 16 models in September.

Now let's take a look at Western models during the same period. Major US frontier labs such as OpenAI, Anthropic, Google, SpaceXAI, and Meta launched or updated at least 10-12 models, variants, and specialized reasoning sub-tiers in September.

US labs such as Anthropic have recently issued warnings over safety guardrails and cyber-risks associated with open-weight releases, especially those from China. In fact, US labs and even government representatives have accused China of distilling US frontier models to release open-weight models. 

Simply put, frontier models are the most advanced AI systems currently available, while open-weight models give developers access to the model's underlying weights, making it possible to customise and run them independently rather than through the company's own service. And AI distillation, also known as knowledge distillation, is a technique in which a large, complex and computationally expensive model, known as the "teacher", trains a smaller and faster model.

To be sure, open-weight releases are not exclusive to Beijing, US giants like Meta, along with open-weight variants from OpenAI and French lab Mistral, have made high-capability models freely available to global developers.

Anthropic and global security researchers raise valid concerns regarding automated exploit generation, guardrail failures, and cross-border model distillation. But safety dilemmas are not isolated to Chinese developers only; they are inherent to open-weight models worldwide. The tension between rapid deployment and safety verification affects the entire tech industry, not just Beijing.

The biggest takeaway perhaps is that in the global AI arms race, speed and cost-efficiency dictate survival. So until global governance frameworks establish binding standards for both open-source and proprietary models, fast-tracked model drops will likely remain the baseline for both China and the US.