Stanford University's Institute for Human-Centered Artificial Intelligence (HAI) released its 2026 AI Index Report today, warning that responsible AI development is failing to keep pace with advancing capabilities.
The report highlights a significant rise in real-world harms, with documented AI incidents climbing from 233 in 2024 to 362 in 2025. This surge follows a period of unprecedented growth, with AI reaching 53 percent of the global population in just three years.
Researchers noted that while organizations report an 88 percent adoption rate, safety benchmarks are lagging. "Responsible AI is not keeping pace with AI capability, with safety benchmarks lagging and incidents rising sharply," the report states.
Performance gaps and reliability issues
Despite improvements in coding proficiency—where success rates on the SWE-bench test jumped from 60 percent to nearly 100 percent in one year—reliability remains inconsistent. The AA-Omniscient Index shows hallucination rates across 26 studied models ranging from 22 percent to a staggering 94 percent.
These errors have already led to legal repercussions. The report cites a case where attorneys used AI models to generate over two dozen fake citations and misrepresentations of fact, resulting in a reprimand from the US Sixth Circuit Court of Appeals.
Even advanced models struggle with basic reasoning tasks. OpenAI’s GPT-5.4 High correctly identified analog clocks only 50.6 percent of the time in ClockBench testing, significantly lower than the 90 percent accuracy achieved by unspecialized humans.
Physical automation also faces steep hurdles. According to the BEHAVIOR-1K simulation benchmark, robots succeeded in only 12 percent of tested household tasks.
