Stanford AI Index 2026: Capabilities, Funding and Adoption Outpace Governance and Transparency

The latest arXiv update of Stanford University’s closely watched AI Index reinforces the report’s central warning: artificial intelligence is getting more capable, more widely used and more heavily funded faster than the systems designed to evaluate, govern and track it. Stanford HAI, the university’s Institute for Human-Centered Artificial Intelligence, says documented AI incidents rose to 362 in 2025 from 233 a year earlier, while U.S. private AI investment hit $285.9 billion and industry produced more than 90% of notable frontier models in 2025.

That mismatch is the headline finding of the “Artificial Intelligence Index Report 2026,” which Stanford HAI first released publicly on April 13 and which was later posted to arXiv on April 14. A revised version appeared on arXiv on June 25. The report is the ninth edition of the AI Index, an annual benchmark project published since 2017 as an independent initiative of Stanford HAI and widely used by policymakers, researchers, companies and news organizations to track the state of AI.

The arXiv abstract opens, “Welcome to the ninth edition of the AI Index report.” The first author listed on the arXiv version is Sha Sajadieh, AI Index lead at Stanford HAI. In Stanford HAI’s own summary of the findings, the institute put the core message more bluntly: “This year’s AI Index report reveals AI’s capabilities are advancing quickly; less so, our ability to measure and manage them.”

On performance, the report argues that “AI capability is not plateauing. It is accelerating.” It says several leading systems now meet or exceed human baselines on PhD-level science questions, multimodal reasoning — meaning tasks that combine text, images and other formats — and competition mathematics. It also points to rapid gains on coding-related evaluation: performance on SWE-bench Verified, a benchmark for software engineering tasks, climbed from about 60% to nearly 100% in a year.

The report also says the race at the top is tightening internationally. According to Stanford HAI, the model-performance gap between the United States and China has “effectively closed,” with models from each country trading the lead since early 2025. As of March 2026, the report says, Anthropic’s top model held a lead of 2.7%.

Money continues to pour in. Stanford HAI says U.S. private AI investment reached $285.9 billion in 2025, compared with $12.4 billion in China, while cautioning that private-investment comparisons may understate China’s government-directed AI spending. Globally, the report says corporate AI investment reached $581.7 billion in 2025, including $344.7 billion in private investment.

Consumer uptake is also moving quickly. The report says generative AI reached 53% population adoption within three years, faster than the personal computer or the internet over comparable periods. It places U.S. adoption at 28.3%, ranking 24th, and estimates the annual value of generative AI tools to U.S. consumers at $172 billion by early 2026.

But the report’s broader argument is that the institutions around AI are not keeping up. Stanford HAI says documented AI incidents climbed sharply in 2025. At the same time, the average score on the Foundation Model Transparency Index fell to 40 from 58, suggesting less public visibility into how major AI systems are built and deployed. The report links those trends to a wider shortfall in governance frameworks, evaluation methods, education systems and the data infrastructure needed to understand AI’s effects.

It also highlights how concentrated the underlying infrastructure has become. The United States hosts 5,427 data centers, the report says, more than 10 times as many as any other country. And Taiwan Semiconductor Manufacturing Co., or TSMC, fabricates almost every leading AI chip, underscoring how much cutting-edge AI depends on a narrow physical supply chain.

This year’s edition adds several new sections that reflect AI’s widening reach. Stanford HAI introduced standalone chapters on AI in science and AI in medicine, added new estimates of generative AI’s economic value and labor-market effects, and included an analytical framework for thinking about AI sovereignty. The science chapter was developed in collaboration with Schmidt Sciences.

Stanford HAI has made the full report and public data available online, with the complete document also archived on arXiv. Taken together, the 2026 index presents a simple but increasingly consequential picture: AI systems are improving and spreading fast, while the tools to measure, oversee and explain them are falling behind.