A Chinese open-source model, GLM-5.2, stunned Silicon Valley this weekend by matching the agentic capabilities of Anthropic's Opus 4.8, according to Axios. The development comes as Five Eyes leaders warned Monday the world is months away from AI models that dramatically accelerate cyber threats.
Yet preparations are stalling. The Trump administration remains locked in debate over how to safely release Anthropic's Fable 5 and Mythos 5 models, with national security concerns pitting internal factions against each other. Industry-wide confusion over how to measure AI risk compounds the paralysis.
Stanford's AI Index Report indicates Chinese models have largely erased the U.S. quality advantage over the past year. Former White House AI czar David Sacks stated this month that the U.S. only has a slight lead, though the timeline for China fully closing the gap remains hotly contested.
The core disagreement among security experts is no longer whether China is catching up, but how quickly. With open-source models like GLM-5.2 proliferating, the window for coordinated safety measures may be narrowing faster than regulators can act.
Some analysts caution that benchmarks may overstate Chinese progress, noting that production-grade safety testing and deployment scale still favor U.S. frontier labs. The true gap, they argue, may be wider than the raw capability scores suggest.