Mythos Comes Back But Not for Everyone
Summary
This episode analyzes the US government’s ad hoc licensing regime for frontier AI, detailing the limited release of Anthropic’s Claude Opus 4.1 and OpenAI’s GPT 5.6 to select partners. It covers industry backlash against these arbitrary access restrictions and discusses how companies like Coinbase are pivoting to Chinese open-weight models to cut costs. The host concludes that while current delays may be temporary, this precedent establishes a lasting geopolitical advantage for the US by controlling access to the most advanced AI capabilities.
Topics discussed
Intro and sponsor announcements
Anthropic's Mythos return and government licensing
OpenAI's GPT-5.6 limited release and government access
GPT-5.6 model variants, pricing, and benchmark performance
METER evaluation and community skepticism on benchmarks
Public reaction and the 'vibe shift' in AI community
Analysis of US AI policy, regulation, and global competition
Sponsors: KPMG, Pencils, Mission Cloud, and OutSystems
Critique of arbitrary policy and lack of clear framework
Chinese AI models, cybersecurity claims, and GLM 5.2
Shift to Chinese models due to cost and access restrictions
Long-term implications: US intelligence advantage and future access
Listen ad-free on Castria