Licensing and frontier AI, in plain English
Seven sourced guides to both meanings of AI licensing: the government permits, filings and disclosure rules that apply to the most capable models, and the private licenses that govern model weights and training data. Each guide states its facts as of October 2026 and links to the primary sources. Sources are listed at the bottom of every guide.
All guides
AI licensing explained
“Licensing” can mean a government permit, a filing, a notice, a transparency duty, an export license, or a private contract on weights and data. This guide sorts them out and traces the debate back to OpenAI’s May 2023 Senate testimony.
Frontier AI laws compared
A plain comparison of the binding rules for the largest AI models, with thresholds, duties, dates and penalties. It also covers the December 2025 executive order targeting state AI laws and where federal preemption stands.
Compute thresholds
Most frontier rules decide coverage by training compute, using 10^25 or 10^26 operations. This guide shows the two estimation methods with worked examples and traces the 10^26 number from the revoked 2023 executive order into state law.
AI export controls
Explains the 2022 chip controls, the January 2025 AI Diffusion Rule and its unfinished rescission, and the January 2026 case-by-case policy for Nvidia H200 exports to China. Includes a dated status table.
Open-weight model licenses
Downloadable weights do not make a model open source. This guide compares the main model licenses on commercial use, use restrictions, user caps and attribution, and explains the OSI Open Source AI Definition 1.0.
Training data licensing
A table of AI content licensing deals with the reported dollar figures and where each comes from, followed by the status of Bartz v. Anthropic, Kadrey v. Meta, Thomson Reuters v. Ross, NYT v. OpenAI, Getty v. Stability and GEMA v. OpenAI as of October 2026.
Should superintelligence need a license?
Sets out who has proposed licensing frontier or superintelligent AI, from OpenAI and Senate sponsors to the UN Secretary-General, and the strongest objections from open-source advocates, NTIA and critics of compute thresholds. Takes no side.