Stanford index flags trust and compute gaps

Stanford’s 2026 Artificial Intelligence Index says the technology is getting better faster than the systems meant to measure, regulate, and explain it. (hai.stanford.edu) The report was released April 13 by Stanford University’s Institute for Human-Centered Artificial Intelligence, which has published the index since 2017 and says this edition runs more than 400 pages. (hai.stanford.edu) Stanford says global opinion is split: 59% of respondents in 2025 said Artificial Intelligence products offer more benefits than drawbacks, up from 55% in 2024, while 52% also said those products make them nervous. (hai.stanford.edu) The sharpest divide is between specialists and the public in the United States. On jobs, 73% of experts expected a positive effect from Artificial Intelligence, compared with 23% of the public; on the economy, the gap was 69% to 21%. (hai.stanford.edu) Americans also showed deep skepticism about oversight. Stanford says the United States had the lowest trust in its own government to regulate Artificial Intelligence responsibly among countries surveyed, at 31%, versus a 54% global average. (hai.stanford.edu) The index ties some of that strain to the scale of the buildout behind modern systems. Stanford estimates global Artificial Intelligence compute capacity has grown more than threefold a year since 2022, reaching a 30-fold increase since 2021. (spectrum.ieee.org) That buildout now carries visible resource costs. Stanford estimates Grok 4 training emissions at 72,816 tons of carbon dioxide equivalent, says Artificial Intelligence data center power capacity reached 29.6 gigawatts, and says annual GPT-4o inference water use may exceed the drinking water needs of 12 million people. (hai.stanford.edu) Stanford also says the information needed to judge those systems is getting harder to obtain. Its Responsible Artificial Intelligence chapter says the average Foundation Model Transparency Index score fell to 40 in 2025 from 58 in 2024, with major disclosure gaps around training data, compute, and post-deployment effects. (hai.stanford.edu) The report places those trust and transparency gaps alongside a tightening race between the United States and China. Stanford says United States and Chinese models traded the top spot several times since early 2025, and as of March 2026 Anthropic’s leading model was ahead by 2.7%. (hai.stanford.edu) The United States still led notable model releases in 2025, with 50, according to IEEE Spectrum’s summary of Stanford’s data, while China led industrial robot installations with 295,000 in 2024, compared with about 34,200 in the United States. (spectrum.ieee.org) Sha Sajadieh, who leads the index, said the central question is whether society is prepared to absorb the disruption and decide how to use the technology. Stanford’s data suggests the machines are scaling faster than public confidence in the people building and governing them. (hai.stanford.edu)