← All stories
Geopolitics

Brad Gerstner confirms GLM 5.2 contains Mythos watermarks, evidence of Chinese model distillation

All-In Podcast · More Trillion Dollar IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts · July 11, 2026
Brad Gerstner confirms GLM 5.2 contains Mythos watermarks, evidence of Chinese model distillation
All-In Podcast
All-In Podcast
More Trillion Dollar IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts
"I will tell you, GLM 5.2 has watermarks from Mythos all over it, right? So we know they were distilling."
Brad Gerstner stated that Chinese AI model GLM 5.2 contains detectable watermarks from Mythos, providing evidence that Chinese labs are distilling American frontier models. He indicated the U.S. government plans to take steps against distillation practices. This comes amid broader U.S.-China AI competition, with Washington focused on maintaining technological superiority over Beijing.

About this episode

Jason Calacanis, David Sacks, and Brad Gerstner discuss the explosive growth of frontier AI labs and the historic launch of Trump Accounts, the government investment platform giving every American child an ownership stake in the economy. Chamath Palihapitiya joins briefly before heading to enterprise software sales calls. The episode centers on two major storylines: the unprecedented revenue growth at OpenAI and Anthropic as they prepare for potential IPOs at over $1 trillion valuations, and Brad Gerstner's announcement that Trump Accounts signed up 1.5 million users and received over $1 billion in deposits within 24 hours of its July 4th launch. Gerstner reveals President Trump has ordered automatic account creation for all 70 million American children, while Michael Dell committed over $6 billion and SpaceX's Gwynne Shotwell pledged $350 million to fund accounts for lower-income kids. On the AI front, Gerstner projects Anthropic could reach $100 billion in revenue and grow 3-5x from there, representing $200 billion in single-year revenue growth that would be unprecedented in business history. However, Chamath warns his company's AI token costs are doubling every 45 days while productivity gains remain flat at 5%, questioning the sustainability of enterprise AI spending. CTOs from Uber, DoorDash, and Databricks shared detailed strategies for optimizing AI costs through intelligent model routing and custom harnesses. The discussion also covers China's reported consideration of restricting overseas access to Chinese AI models as a national security measure, with Gerstner confirming Chinese model GLM 5.2 contains watermarks from distilled American models. The besties debate whether frontier model intelligence will converge with open source alternatives or whether the gap will expand as superintelligence becomes recursive.

Key takeaways

More stories More from All-In Podcast