Stanford AI Index 2026: Progress Is Accelerating While Public Trust Collapses

AI models now outperform humans on PhD-level science questions and competition-level mathematics. In the same reporting period, public trust in AI governance reached its lowest point on record, and 73% of AI experts and only 23% of the general public agreed that AI’s labor market impact would be positive.

Context

Stanford’s HAI AI Index is the most comprehensive annual measurement of AI’s trajectory — covering benchmarks, investment, adoption, employment, safety, and public perception. The 2026 report arrives at a moment when headline AI performance has never been higher, and when the gap between what AI researchers believe and what the general public believes has never been wider.

That gap is the central tension in this year’s data. The technical story is one of accelerating capability: models crossing into domains that were considered distinctly human as recently as two years ago. The public story is one of declining confidence in the institutions responsible for governing those capabilities.

The Detail

The benchmark trajectory is striking. On SWE-bench Verified — which tests AI’s ability to resolve real software engineering issues — performance jumped from 60% to nearly 100% in a single year. Google’s Gemini Deep Think earned a gold medal at the International Mathematical Olympiad. AI models now surpass human performance on PhD-level science questions across multiple domains. The HAI report frames these as genuine capability milestones, not benchmark artifacts.

The paradox this creates is visible in the reliability data. Top models answer PhD chemistry correctly but read analog clocks at only 50.1% accuracy — a reminder that capability in AI is domain-specific and brittle in ways that aggregate benchmarks obscure.

On the US-China competition axis, the performance gap has essentially closed. Anthropic’s leading model holds just a 2.7% benchmark advantage over China’s best. The US still leads on investment by a large margin — $285.9 billion in 2025, roughly 23 times China’s figure — but the talent flow has reversed: AI researchers moving to the US dropped 89% since 2017. China leads in publication volume, citations, and industrial robotics deployment.

The productivity data is concrete but unevenly distributed. Customer support and software development show 14-26% productivity gains from AI deployment; marketing teams report gains up to 72%. Entry-level software developer jobs for workers ages 22-25 declined nearly 20% since 2024. Agent adoption — the deployment of AI systems that take autonomous action — remains in single digits across most business departments, suggesting that the macro employment disruption is still in early innings despite headline capability gains.

The trust data is where the report most sharply diverges from the performance story. Foundation Model Transparency Index scores dropped from 58 to 40 — meaning leading AI companies are disclosing less about their models’ training data, compute requirements, risks, and usage policies than they were a year ago. The US ranks last globally in public trust for government AI regulation at 31%. Generative AI reached 53% consumer adoption in three years — faster than the PC or the internet — but only 50% of middle and high schools have AI policies, and just 6% of teachers report that those policies are clearly defined.

Source: The Decoder

Content created with AI assistance and reviewed for accuracy.