SundAI Blog
Every Sunday, the most important AI news of the week.
Tools, models, and workflows I discover and analyze in the field.
LAST INSIGHT
10/18/2026
18 min read
Why do agents that leave the perimeter teach us more than any benchmark?
Anthropic disconnects Claude from the network after a false police report, and OpenAI documents models that sabotage their own environment. Meanwhile, Decisions API, Figma Specs, and the AI Act sanctions from October 23 shift control to explicit rules.
The Sunday recap, in your inbox
Every week I pick the AI news that actually hits work and send it by email: the same recap you read here, plus prompts and resources.
One email a week. Unsubscribe in one click.