$40 Trillion in Debt and the Only Exit Is Inflation, The Guy Who Wrote OpenAI's Safety Rules Just Quit, The Situation in France Is Only Getting Worse| Weekly Recap

Welcome to your Impact Theory Weekly Recap!Trump told Time that certain levels of inflation will pay off the national debt "very rapidly," and two days later his own National Economic Council director Kevin Hassett was asked on Bloomberg whether the White House planned to inflate its way out of the debt and answered "no, absolutely not." The team starts there and works outward into why inflation is the historically normal exit from a $40 trillion debt load, why the debt is never actually going to be repaid, and why being the world's reserve currency makes running permanent deficits something closer to a job requirement than a choice. That framing explains the nationalist argument for giving up reserve status, which would let the US weaken the dollar, compete on trade and rebuild manufacturing, at the cost of the exorbitant privilege that funds everything else. The core warning is that inflating away debt only works if growth and real wages keep pace, and that without it you get the asset-owner split widening into something the team argues ends badly. The back half covers the 1971 break from gold, the hollowing out of manufacturing, a question about how much of the debt traces to tax cuts rather than spending, and what Kevin Warsh might actually be trying to accomplish.OpenAI lost four safety staff in a single week, three fired for sharing information with an outside safety group and one who resigned and went public. David Robinson spent three and a half years at the company, led the writing of the safety reports that shipped with twelve frontier model launches, and drafted the current Preparedness Framework before quitting and publishing an essay in The Atlantic titled "I Quit OpenAI Because Its Culture Is Broken." The team works through why this is happening, arguing the pressure is structural: OpenAI's revenue is tied entirely to the AI buildout while competitors like Google and Meta have cash cows that aren't, which makes slowing down far more expensive for one than the other. From there the conversation turns to what containment should actually look like, including Nvidia's approach of dropping models into permission structures with adversarial AI watching them, why you don't need to believe in Terminator to worry about a model spinning up copies and hiding code, and why handing this to government regulators is the wrong answer when existing liability law already has teeth. It closes on who gets to watch the watchers, why effective altruism as a govern