US-China AI gap narrows as safety concerns grow

Increasingly, the US–China AI gap is narrowing, challenging assumptions of a sustained US lead. According to the 2026 AI Index from Stanford’s Human-Centred AI Institute, top models from both countries have traded leadership since 2025. As a result, the performance gap between leading models has shrunk to just 2.7%.

While the US still produces more frontier models and holds stronger patent influence, China has gained ground in research output, citation share, and patent volume. This indicates a shift from dominance to competition, where leadership changes with each new model release rather than remaining structurally fixed. 

At the same time, the report highlights a critical imbalance in AI development: safety and governance are not keeping pace with capabilities. Most frontier models report performance benchmarks consistently, but responsible AI metrics (covering fairness, security, and human oversight) remain largely unreported. This lack of standard evaluation makes it difficult to compare risks across systems. 

The consequences are already visible. Documented AI incidents rose significantly, reaching 362 in 2025, while organisations reporting strong incident response capabilities declined. This suggests that as AI adoption accelerates, governance maturity is falling behind. 

• US and China now compete closely in AI model performance.  

• China leads in research output and patent growth.  

• AI safety benchmarking remains inconsistent and limited.  

• AI incidents are increasing while governance readiness declines.  

Public perception adds another layer of complexity. While global optimism about AI is rising, so is anxiety, reflecting growing uncertainty about its long-term impact. Trust in regulation varies significantly across regions, influencing how AI policies will evolve. 

Ultimately, the closing US-China AI gap signals a more competitive global landscape. In 2026, the defining challenge is no longer who leads in capability, but who can build trustworthy, safe, and well-governed AI systems at scale. 

 

Source: 

https://www.artificialintelligence-news.com/news/ai-safety-benchmarks-stanford-hai-2026-report/