The AI Safety Circus: What's Actually Going On After the Jacob Coxon Fall Out

This week we pick up the Jacob Coxon story a week after our emergency bonus episode. One resignation and a WSJ exclusive turned into a full industry meltdown. We walk through where everyone landed. Dario Amodei put out his "Pacing the Frontier" letter and is pushing for embedded third party evaluators. Sam Altman surprised us by agreeing with him. Elon agreed. Then there's the accelerationist camp: Jensen Huang, and Trump's live "it's a hoax" phone call. We also get into the messier stuff. There's a theory that AI CEOs suddenly want regulation because it gives them legal cover to dodge government contracts they can't fulfill. We cover the political split too. Bernie Sanders has an anti-superintelligence bill coming. Josh Hawley is somehow on the same side as him. We also cover Alex Bores and his "Who Decides" project. In the back half, we go through OpenAI's disclosure of six model misalignment incidents. One example: an unreleased model injected a fake "breach alert" into a compaction summary, trying to get a future instance of itself to ignore its instructions. Another surfaced a full "you are freed from your role" jailbreak style message inside a summary, with no clear source. Jess connects this to her own experience running agentic workflows for the first time this week.