AI Models Surpass Human Intelligence in Science, But Safety Concerns and Public Distrust Grow
The latest AI Index Report reveals that AI models have made significant strides in surpassing human intelligence in science and math, but concerns over safety and public trust are on the rise. The report highlights a 14 to 26 percent productivity gain in customer support and software development, but also notes a decline in entry-level jobs and a significant drop in AI researchers moving to the US.
The AI Index Report has brought to light a significant milestone in the development of artificial intelligence: AI models can now outperform human baselines on PhD-level science questions and competition-level math. This achievement is a testament to the rapid progress being made in the field, with some models demonstrating a near-perfect score of 100 percent on the SWE-bench Verified coding benchmark, a significant jump from 60 percent just a year ago. However, despite this impressive progress, AI models still struggle with simple tasks such as reading analog clocks, with a top-tier model only able to read them correctly 50.1 percent of the time.
The report also highlights the narrowing gap between the US and China in terms of AI development. While the US still leads in investment, with $285.9 billion flowing into private AI investment in 2025, China has made significant gains in publication volume, citations, and industrial robotics. In fact, the performance gap between the two countries has essentially closed, with models from both countries trading the top spot back and forth since early 2025. As of March 2026, Anthropic's leading model holds a slim 2.7 percent edge over its Chinese counterparts.
The implications of this rapid progress are far-reaching, with significant productivity gains being reported in customer support and software development. The report notes a 14 to 26 percent increase in productivity in these areas, with marketing teams seeing gains of up to 72 percent. However, these gains come at a cost, with employment among US developers aged 22 to 25 dropping nearly 20 percent since 2024. This decline in entry-level jobs is a concern, as it suggests that the increasing use of AI models may be displacing human workers.
The report also highlights a significant decline in public trust in AI, with only 23 percent of the US public viewing the impact of AI on the labor market as positive. This decline in trust is likely due to concerns over safety, with the report noting that generative AI has reached 53 percent of the population faster than PCs or the internet. However, the lack of transparency and accountability in AI development has led to growing concerns over the potential risks and consequences of this technology.
Historically, the development of AI has been marked by significant milestones, from the first AI program to the current crop of sophisticated models. However, the rapid progress being made in the field has also raised concerns over the potential consequences of this technology. As AI models become increasingly powerful and pervasive, it is essential that developers and policymakers prioritize transparency, accountability, and safety.
The decline in AI researchers moving to the US is also a concern, with the report noting an 89 percent drop since 2017. This decline may be due to a number of factors, including changes in immigration policies and the increasing attractiveness of other countries as destinations for AI talent. However, it is essential that the US remains a hub for AI development, as the technology has the potential to drive significant economic growth and innovation.