The AI race has no brakes: 23 policy ideas to build them before it's too late

Policy experts at IFP just dropped a menu of 23 specific interventions for governments trying to wrap their heads around recursive self-improvement (RSI) in AI. The core problem is chillingly simple: we’re barreling forward with only an accelerator pedal. These proposals aim to install the brake pedal, the dashboard, and maybe even a seatbelt.

The recommendations span seven categories, from forcing transparency into automated AI R&D to accelerating defensive applications and building out state capacity to actually understand what’s happening. One intriguing thread: deliberately accelerating the diffusion of AI capabilities toward inference and applications, while simultaneously pouring resources into verification tech and resilience. It’s a hedge — spread the benefits widely so the risks don’t concentrate in a few hands.

What makes this more than just another policy paper is its framing around “low-regret” moves. These aren’t calls to halt progress; they’re about giving the US (and other governments) more optionality on the gameboard. The thinking is brutally pragmatic: the fewer levers you have when a crisis hits, the worse the outcome.

Meanwhile, a separate paper from MIT and Columbia researchers digs into whether AI firms could ever coordinate a slowdown voluntarily. Their game theory model, Racing to Ruin, identifies two non-negotiable ingredients: transparency about technological progress and a baseline assumption that rivals are rational actors. The grim twist? Even with both, firms face a perverse incentive to stop second. The first mover to pause gambles that their competitor will reciprocate — and the speed of information flow determines whether that gamble pays off. Faster news makes simultaneous coordination harder, because firms can wait and verify before committing. I’ve been covering this space long enough to know that voluntary slowdowns without binding verification mechanisms have a near-perfect track record of failure. These papers together paint a picture of a race that won’t self-correct without external guardrails.