This study is considered essential reading for policymakers in academia, business, and politics. However, as transparency among top AI developers decreases, Sajadier acknowledged that it is becoming harder to know what needs to be addressed, particularly in regulation and legislation, “to understand what risks we want to mitigate as a society in the first place.”
“With all the enthusiasm and evangelism around AI, consideration of how to responsibly manage its applications and use cases has taken a backseat,” Stephen Bater, executive director of the East Bay Economic Development Alliance, wrote in KQED.
He observed that jobs tied to the physical world appear to have the lowest risk of disruption, especially in fields such as construction, health care, and public safety. But he has concerns beyond AI’s direct impact on the labor market. “There has been strong respect for delaying or ignoring sensitive core human rights and quality of life issues related to personal privacy, safety and security.”
Other critics of AI go further. “What is at odds with leading experts and the general public are the companies themselves, who are racing to replace humans as quickly as possible,” Chase Hardin, a spokesperson for the Future of Life Institute, a nonprofit dedicated to reducing global catastrophic and existential risks from transformative technologies, said in an email.
Hardin said public opinion polls are clearly negative about the risks of AI. “You can argue about the reasons for that, but the public is deeply skeptical of business itself, deeply skeptical of technology, and deeply concerned about what that means for our children.”
Key takeaways from the AI Index report include:
1. AI experts and the general public have very different perspectives on the future of the technology. Assessing the impact of AI on employment, 73% of U.S. AI professionals said the technology will have a positive impact on employment, compared to just 23% of the general public, a 50-point difference. Similar rifts have emerged in the economic and medical sectors.
Trust in governments regulating AI varies globally. Among the countries surveyed, the United States had the lowest level of trust in their government regulating AI, at 31%. Globally, the EU is trusted more than the US or China to effectively regulate AI.
2. AI capabilities are accelerating and impacting more people than ever before. By 2025, private companies will have built more than 9 out of 10 of the world’s most powerful AI models, and some of those models are now outperforming human experts on PhD-level science and advanced math exams.
3. Productivity gains from AI are emerging in many of the same sectors where entry-level employment is beginning to decline. Studies have shown productivity increases of 14% to 26% in customer support and software development, but weaker or negative effects were seen in tasks that require more judgment.
In software development, where the productivity gains from AI are most evident, employment of U.S. developers between the ages of 22 and 25 has declined by nearly 20% since 2024, even as the workforce of older developers continues to grow.
4. Students are using AI, but educational institutions are still playing catch-up. Four out of five U.S. high school and college students are currently using AI in their schoolwork, but only half of middle and high schools have an AI policy in place, and only 6% of teachers say their policies are clear.
5. AI is transforming clinical care, but rigorous evidence remains limited. AI tools that automatically generate clinical notes from patient encounters were significantly introduced in 2025. Across multiple hospital systems, physicians reported up to an 83% reduction in time spent writing notes and significantly reduced burnout.
However, outside of specific tools, the evidence base for clinical AI remains thin. A review of more than 500 clinical AI studies found that nearly half relied on exam-style questions rather than actual patient data, and only 5% used actual clinical data.
6. AI’s environmental footprint is growing along with its capabilities. Last year, training a single AI model produced approximately the same amount of carbon dioxide as 16,000 round-trip flights from San Francisco to New York. Researchers estimate that just one run of the widely used AI model GPT-4o could consume enough water for a year to meet the drinking needs of everyone in Los Angeles and San Francisco combined.
7. The United States leads the world in AI investment, but its ability to attract global talent is declining. US private AI investment will reach $285.9 billion in 2025, more than 23 times the $12.4 billion invested in China, but looking at private investment numbers alone likely underestimates China’s total AI spending when government guidance funds are taken into account.
The US is also leading the way in entrepreneurial activity, with 1,953 new AI companies receiving funding in 2025, more than 10 times as many as the next closest country, the UK. However, the number of AI researchers and developers immigrating to the United States has declined by 89% since 2017, and by 80% in the last year alone.
8. The performance gap between US and Chinese AI models has effectively narrowed. The US and Chinese models have swapped lead multiple times since early 2025.
The United States is still building more of the world’s most powerful AI models, while China is publishing more research, filing more patents, and putting more robots into its factories. South Korea stands out for its density of innovation, leading the world in the number of AI patents per capita.
