September 2026

9 research entries this month.

2026-09-20Safety & Risk

The Regulatory Floor Is Rising State by State, Faster Than Federal Law Can Follow

Five more states passed AI laws this year while Washington still works from a March policy framework, not a statute -- and Gartner expects over half of large enterprises to face mandatory AI compliance audits this year.

2026-09-20Cost & Infrastructure

AWS and Nvidia's 2-Million-GPU Bet, and the Bottleneck Neither Can Buy Its Way Past

AWS is adding 2 million more Blackwell Ultra and Rubin GPUs on top of an already-raised 1-million-GPU commitment -- but the packaging step that turns a chip into a shippable accelerator is sold out into 2027.

2026-09-20Research Trends

Reasoning Models Hallucinate More, Not Less -- and Extended Thinking Is the One Lever That Reliably Helps

A counterintuitive 2026 finding across multiple benchmark studies: the models marketed as most capable, because they reason step by step, are measurably less reliable on plain factual questions than models without that chain-of-thought process.

2026-09-19Enterprise Adoption

Agentic AI in 2026: Wide Adoption, Thin Production

Nearly 8 in 10 enterprises report using AI agents -- but fewer than 1 in 10 have scaled past a pilot. The gap between 'we tried it' and 'it runs the business' is this year's real story.

2026-09-18Developer Tooling

The Coding Agent Market Reshuffles: Claude Code Overtakes, Codex Surges, Cursor Cools

JetBrains' latest developer survey shows the biggest swing yet in which AI coding tool developers actually reach for day to day -- and it isn't the tool that had the market a year ago.

2026-09-17Cost & Infrastructure

Inference Costs Keep Falling ~10x a Year — What That Actually Changes

The per-token cost of a given capability level keeps dropping by roughly an order of magnitude annually. That's not just a pricing story -- it resets which architecture choices are still worth making.

2026-09-16Model Releases

September 2026's Model Releases, and What Actually Changed

Four frontier labs shipped in the same month. A rundown of what each release actually claims to improve, not just that it shipped.

2026-09-15Safety & Risk

AI Agents Are Getting Used for Attacks Too

Anthropic documented real, in-the-wild AI-assisted cyberattacks, scams, and surveillance spanning last December through August. The same agentic capability that helps a support team also helps whoever's attacking it.

2026-09-14Enterprise Adoption

Why 40% of Agentic AI Projects Are Expected to Be Cancelled by 2027

Gartner's cancellation forecast isn't a prediction that agentic AI fails -- it's a prediction that most current projects were scoped and governed like a chatbot pilot, not a production system.