Stanford 2026 AI Index: AI Capability Accelerating, Safety Lagging Behind

Stanford HAI's 2026 AI Index Report (400+ pages) finds AI capability accelerating rapidly: SWE-bench Verified rose from 60% to near 100% in one year, and Humanity's Last Exam accuracy jumped from 8.8% (o1) to over 50% (Claude Opus 4.6, Gemini 3.1 Pro). Industry produced over 90% of notable frontier models in 2025. Organizational AI adoption reached 88%, and 4 in 5 university students now use generative AI. However, responsible AI lags: documented incidents rose to 362 in 2025 from 233 in 2024, and almost all leading developers report capability benchmark results while responsible AI benchmark reporting remains spotty. Research found improving one responsible AI dimension (e.g., safety) can degrade another (e.g., accuracy).