A Chaotic Week in AI: Bans, Breakthroughs, and the Open Source Surge

The past few weeks have been among the most eventful in recent AI history. The US government forced Anthropic to hold back its Mythos model and then blocked OpenAI from releasing GPT-5.6, even deciding which customers can get access — effectively locking startups out. Meanwhile, a Chinese food delivery company called Z AI open-sourced GLM 5.2, a near-frontier model that the AI community is calling the "open source GPT-3.5 moment." And Anthropic quietly dropped Sonic 5, a faster, cheaper model that undercuts its own Opus line on price while delivering strong agentic coding performance. To make sense of it all, host Alex assembled three AI founders: Victor Perez of Korea AI (creative image generation), Div Garg of AGI Inc. (on-device mobile agents), and Andrew Berman of RunLayer (enterprise agent control planes).

The Regulatory Crackdown and Its Fallout

The US government's decision to force OpenAI to hold back GPT-5.6 and restrict which customers can access it marks a dramatic shift from the free-range AI environment of just months ago. Andrew Berman called the policy "insane," arguing it gives China and other actors room to encroach on US technological supremacy. "Open source models are being distilled from our world-class models," he said, "and we should be pushing ahead as fast as possible."