OpenAI's new dots agents look like plush toys, Google handed its strongest model to cyber defenders before anyone else, and Anthropic is paying to train the engineers who deploy Claude. This episode walks through why the packaging, the access rules, and the training money all say something about who gets to use frontier AI and on what terms. It also covers OpenAI's apology to Australia, a safety lead's resignation, Castmagic for repurposing podcasts, and a step-back prompt you can paste into any chatbot.

What's Covered

* OpenAI launches dots, plush-shaped agents that book flights and update code

* Google gives Gemini 4 Argon to its Fairwind Program of cyber defenders first

* Anthropic puts $100 million into Claude Frontier Academy

* Other headlines, including OpenAI's apology to Australia and a safety lead's resignation

* Tool of the Week: Castmagic

* Prompt of the Week: step-back prompting

Sources

* SF Standard: OpenAI launches dots

* CNBC: Google Argon and the frontier race

* The Hacker News: Google rolls out Gemini 4 Argon

* TechWire Asia: Gemini 4 Argon and cybersecurity access

* Anthropic: Claude Frontier Academy

* Castmagic

* arXiv: Take a Step Back, evoking reasoning via abstraction

* TechCrunch: OpenAI apologizes to Australia

* Yahoo Tech: OpenAI safety employee David Robinson resigns