Summary
The All-In Podcast — with Jason Calacanis, David Sacks, Chamath Palihapitiya, and David Friedberg — debates what became one of the most viral AI safety events of 2026: the resignation of Jacob Coxin, a researcher who worked at both OpenAI and then Anthropic (the last six weeks before quitting), publicly claiming both labs are “gambling with our lives” in a race toward self-improving superintelligence. The post reached an estimated 110–150 million views within 24 hours, with Anthropic alignment lead Evan Hubinger amplifying it and stating he personally estimates over a 10% chance AI kills all humans this decade.
David Sacks makes an extended case that the event was an orchestrated campaign: the account had no prior activity, three EA-funded policy organizations amplified it within the first 15 minutes, and a Wall Street Journal story appeared minutes before the post itself — suggesting embargo coordination. The episode attempts a genuine steelman of the recursive self-improvement argument: if AI can automate AI researchers, you get a feedback loop with no human in the loop. The counterargument from the panel is that we can’t yet reliably complete mundane agentic tasks, let alone remove humans from AI development cycles.
The episode also covers OpenAI’s reported Navier-Stokes breakthrough using 10,000 agents and Sam Altman’s use of it to argue for pacing, plus political responses from Bernie Sanders and Governor Pritzker. A sharp, adversarial four-way debate for anyone tracking the AI safety discourse and its intersection with politics and media.
📺 Source: All-In Podcast · Published September 12, 2026
🏷️ Format: Podcast
