← All Issues
Sunday, September 20, 2026

AI Daily

Your Automated Intelligence Briefing

AI Models

The New AI Models Refuse to Talk. That Is the Whole Point.

Two releases this week do not output words at all. They hand back a probability. It makes them cheap, fast, and unable to hallucinate, and their makers think this "System One" model is the start of a whole new class of AI that spreads intelligence into everything the way the early internet did.

Read full story →
Geopolitics

Neither Side Will Slow Down. This Week They Started Talking Anyway.

Ahead of a Trump-Xi summit, US and Chinese officials met to discuss AI safety without slowing the race, under a fresh advisory accusing six Chinese firms of copying US models at industrial scale. The realistic goal is not trust. It is a phone line for when something goes badly wrong.

Read full story →
Media & Copyright

They Called It a Theft of Unprecedented Proportions. They Were the Ones Doing It.

Newly unsealed filings in the New York Times case show OpenAI and Microsoft staff privately called their own training data "the largest theft of labor in human history" and an "existential threat" to journalism, even as executives chased the "gazillions" and their lawyers argued fair use.

Read full story →