The Debate Over Slowing the AI Race: When to Hit the Brakes on Frontier Models

AI safety analyst and Substack author Zvi Mowshowitz has published a breakdown of the debate over "pacing the frontier" — deliberately slowing the release of the most powerful AI models. The trigger was an open letter urging leading AI industry players to prepare for the possible introduction of such a regime.

What is "pacing the frontier"

"Pacing the frontier" is an AI safety term. Leading labs, either voluntarily or under pressure from regulators, slow the release of new generations of their most powerful models to give safety teams time to catch up on research.

Importantly, the concept does not imply a full stop. It describes a controlled slowdown with conditions agreed in advance. That nuance is what makes the debate practically relevant in 2026, as frontier systems gain capabilities faster than methods for evaluating them.

Why the letter appeared — and why prepare in advance

The key argument of the letter Mowshowitz analyzes is this: the time to prepare for "pacing" is now, not when the pressure of events has already removed any choice. According to the Substack breakdown, the debate has unfolded around two specific questions.

• When slowing down would be reasonable — that is, which triggers (model capabilities, type of incident, regulatory threshold) should set it off.
• Whether it makes sense to prepare — to agree on mechanisms, procedures and commitments in advance, before the situation turns into a crisis.

"After the letter urging preparation for a possible slowdown of the

frontier, a broad discussion unfolded about when such a step would be reasonable and whether it makes sense to prepare for it," is how Mowshowitz himself sums up the debate.

In practice, this amounts to an attempt to answer the hardest question in AI governance: how to agree on coordinated action among competing labs in different countries without creating an asymmetry in which some slow down and others do not.

What this means

The letter itself, and the scale of the debate that followed, are a signal. "Pacing the frontier" is no longer a fringe idea; it has become a serious item on the AI community's agenda. As Mowshowitz stresses, the question is not whether this is good or bad in the abstract, but whether the industry is ready to make a deliberate decision before circumstances make it on the industry's behalf.