Stanford's 2026 AI Index Report landed this April with a paradox that should keep everyone awake: AI is racing forward in capability at the exact moment our ability to see inside it is collapsing.

On the capability side, the leaps are staggering. AI agents handling real-world terminal tasks jumped from 20% success in 2025 to 77.3% in 2026. Cybersecurity agents went from solving 15% of problems in 2024 to 93% today. The SWE-bench coding benchmark rocketed from 60% to near 100% in a single year. Generative AI reached 53% of the global population faster than the PC or internet ever did.

Now the flip side. The Foundation Model Transparency Index — which measures how much leading AI developers disclose about their models — dropped from 58 points to 40. The most capable models, the report notes, "often disclose the least amount of information." Documented AI incidents rose from 233 to 362 in one year. And while executives expect workforce reductions to accelerate, employment among software developers aged 22–25 has already plunged nearly 20% since 2024 — even as senior developer headcount grows.

The US-China AI race is now a dead heat — Anthropic's best model leads China's top contender by just 2.7% as of March 2026, a gap that has traded hands multiple times since early 2025. Meanwhile, global corporate AI investment hit $581.7 billion, up 130% year-over-year. The money is flooding in faster than the guardrails can be built.

Sources: Stanford HAI — 2026 AI Index Report, Stanford HAI — 12 Takeaways, The Median — 20 Takeaways