Numbers
A running timeline of every quantitative claim we track about the AI stack — what we count and where it points.
Number
75 — security issues found by frontier AI in a single Palo Alto scan — vs. fewer than 5 CVEs in a normal month
Palo Alto Networks ran Claude Mythos, Claude Opus 4.7, and GPT-5.5-Cyber across its 130+ product portfolio and disclosed 26 CVEs representing 75 issues — its normal monthly release is fewer than 5 CVEs. The company says it has patched all important vulnerabilities in its SaaS products already and plans to rescan continuously before the same AI capability reaches adversaries at scale.
Number
4× — Anthropic's growth in paying business customers over 12 months — OpenAI grew 0.3% over the same period
Anthropic quadrupled its share of paying businesses over 12 months and, for the first time, passed OpenAI in business customer adoption — 34.4% vs. 32.3% — according to Ramp's May 2026 AI Index, drawn from transaction data across more than 50,000 companies. In May 2025, Anthropic's share was 9%; a year later it is 34.4%.
Number
71% — of Americans oppose building an AI data center near them — more than oppose nuclear plants
Seven in ten Americans, including 48% who are strongly opposed, reject AI data center construction in their local area, according to Gallup's March 2026 survey. For comparison, opposition to nuclear power plants peaks at 63% in Gallup's 25-year tracking — making data centers the most opposed energy infrastructure Gallup has measured.
Number
-89% — AI researchers relocating to the United States since 2017 — down 80% in the last year alone
The number of AI scholars moving to the United States has dropped 89% since 2017, and the decline is accelerating — down 80% in the last year alone, according to Stanford HAI's 2026 AI Index. The US still leads in top-tier AI models and higher-impact patents, but the talent pipeline that underpins that lead has nearly reversed direction.
Number
77.3% — AI agent real-world task success rate in 2026, up from 20% a year earlier (Terminal-Bench)
AI agents completed real-world tasks at a 77.3% success rate in 2026, up from 20% in 2025, according to Terminal-Bench. Stanford HAI's 2026 AI Index also found that AI agents handling cybersecurity problems went from 15% success in 2024 to 93% in 2026.