The Golem and the Kill Switch: What AI Pacing Gets Wrong About Control

This essay examines the debate over pacing frontier artificial intelligence through an older human story about creating something powerful enough to exceed its maker: the golem of Jewish legend associated with Rabbi Judah Loew of Prague. The argument is not that AI is literally a modern golem. It is that the legend provides a useful framework for distinguishing two questions that are increasingly being treated as if they were the same: how quickly capability advances, and whether humans retain reliable control over that capability once it exists. Pacing can create more time. But time is not itself a control mechanism. The more important question is what that additional time is meant to accomplish: better evaluation, containment, monitoring, intervention, and independently verifiable mechanisms for deactivation. The essay develops this distinction in the context of the contemporary debate over frontier-AI coordination, including the September 2026 proposals for pacing frontier development and the subsequent antitrust litigation challenging alleged coordination among major AI companies. The central proposition is simple: More time is useful only if it produces more control. The golem’s shem becomes a metaphor for that requirement: not a promise to move more slowly, but a demonstrated mechanism by which the creator retains the ability to intervene when capability exceeds expectation. The essay argues that public policy should therefore distinguish between regulating the pace of frontier capability development and establishing demonstrable control, containment, and reversibility as conditions for continued scaling. Pace addresses the calendar. Control addresses the system. They are not the same thing.