How should you slow down AI progress if it becomes necessary?

According to an essay on LessWrong by Felipe Calero-Forero, if artificial intelligence progress ever needs slowing—due to misalignment risks, economic displacement from automation, or capabilities outpacing our ability to adapt—the actual policy levers are far more intricate than simply pausing progress. Calero-Forero proposes a three-part framework. First: a hard cap on how much compute AI companies can spend on research and development, combined with a progressive tax for spending below that cap. Second: a separate limit on individual training runs, as a safeguard against workarounds. These target sudden risks like misalignment. Third: a capability-gated tax on deploying powerful AI systems—essentially metering how fast they roll out to control economic disruption. The key concept is the critical window: a moment where intervention is most leveraged, before major risks fully materialize. Calero-Forero argues slowdown policy must target the specific layer of technology where a risk emerges, rather than imposing blunt universal restrictions.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton