Overview
Stanford University’s Institute for Human-Centered AI released its 2026 AI Index — the most authoritative annual snapshot of where AI actually stands. The headline: AI is advancing faster than society’s ability to understand, govern, or trust it.
1. The US and China Are Nearly Tied
Anthropic currently leads Arena model rankings, followed closely by xAI, Google, and OpenAI. Chinese models lag only modestly. The US has 5,427 data centres — more than 10 times any other country. China leads in research publications, patents, and robotics. Worryingly, top labs no longer disclose training data or model parameters — making independent safety research significantly harder.
2. No Plateau in Sight
SWE-bench software engineering scores jumped from around 60% in 2024 to nearly 100% in 2025. AI now meets or exceeds human experts on PhD-level benchmarks. One Stanford researcher put it simply: “I am stunned that this technology continues to improve, and it’s just not plateauing in any way.”
3. How We Test AI Is Broken
One popular maths benchmark has a 42% error rate. Models can be gamed by training on test data. Companies are hiding their responsible-AI benchmark results. For complex technologies like AI agents and robots, benchmarks barely exist yet.
4. AI Is Starting to Affect Jobs
AI adoption is faster than the PC or internet — over half the world now uses AI within three years. Employment for software developers aged 22–25 has fallen nearly 20% since 2022. A third of organisations expect AI to shrink their workforce, particularly in software engineering and supply chain. Productivity gains are real — 14% in customer service, 26% in software development — but concentrated in exactly the fields where entry-level employment is declining.
5. People Have Complicated Feelings
59% of people globally think AI will provide more benefits than drawbacks. But 52% say it makes them nervous. The biggest divide: 73% of AI experts think AI will positively impact jobs versus only 23% of the American public — a 50 percentage-point gap. Americans trust their government least to regulate AI, at just 31%.
6. Governments Are Struggling to Regulate
The EU AI Act’s first prohibitions took effect. The US moved toward deregulation under President Trump, but US states passed a record 150 AI-related bills. California enacted SB 53, mandating safety disclosures and whistleblower protections.
Additional Findings
Education: 4 in 5 US students use AI for schoolwork, but only 6% of teachers say their school’s policies are clear.
Healthcare: AI cut clinical note-writing by up to 83%, but 95% of clinical AI studies lack real patient data.
Talent drain: US private AI investment reached $285.9 billion — 23 times China’s. But AI researchers moving to the US dropped 89% since 2017.
Environmental cost: Training one AI model generates the carbon equivalent of 16,000 round-trip flights from San Francisco to New York. Running GPT-4o may consume enough water annually for all of LA and SF combined.
The Bottom Line
AI is sprinting. The rest of us — governments, educators, employers, and regulators — are still finding our shoes. The gap between expert optimism and public anxiety has never been wider.