According to an annual report from Stanford University, by March, the network handled 66% of computer tasks, approaching a human target of 72%. Just a year earlier, the rate was only 12%. In the specialized programming test SWE-bench Verified, model efficiency soared from 60% to almost 100% over the same period. At the same time, the opportunity gap between US and Chinese development has virtually closed.

Adoption of big language models in companies reached 88%, and global corporate investments in industry for 2025 exceeded $581 billion - double the previous year. But the study warns: control systems and safety standards are hopelessly lagging behind technology. The number of reported AI incidents has increased to 362 from 233 in 2024, and developer reporting on responsible use remains incomplete.

One particular concern is the widespread use of AI in education: more than 80% of American high school and university students apply a neural network for learning, but only half of educational institutions have any rules, and only 6% of teachers consider them clear.