Дискуссия о замедлении ИИ-гонки: когда тормозить разработку фронтирных моделей

AI safety analyst and Substack author Zvi Mowshowitz has published an analysis of the debate around the concept of "pacing the frontier" — the deliberate slowing of the release of the most powerful AI models. The occasion was an open letter calling on leading players in the AI industry to prepare for the possible introduction of such a regime.

What Is "Pacing the Frontier"

"Pacing the frontier" is a term from AI safety: leading laboratories, voluntarily or under pressure from regulators, reduce the pace at which new generations of the most powerful models are released, in order to give safety teams time to catch up in their research.

Importantly, the concept does not imply a complete halt — it refers to a controlled slowdown with pre-agreed conditions. This very nuance makes the debate practically significant in 2026, when the capabilities of frontier systems are advancing faster than the methods for evaluating them.

Why the Letter Appeared — and Why Prepare in Advance

The key argument of the letter that Mowshowitz analyzes is this: preparations for "pacing" must happen now, not when the pressure of circumstances has already taken away the choice. According to the Substack analysis, the debate unfolded around two specific questions.

• When a slowdown would be reasonable — that is, which specific triggers (model capabilities, type of incident, regulatory threshold) should activate it.
• Whether it makes sense to prepare — to coordinate mechanisms, procedures, and commitments in advance, before the situation becomes a crisis.

"Following the letter calling for preparation for a possible slowdown of the frontier, a wide discussion unfolded — when such a step would be reasonable and whether it makes sense to prepare for it,"

Mowshowitz himself formulates the essence of the debate.

In practice, this means attempting to answer the hardest question in AI governance: how to agree on coordinated action among competing laboratories from different countries, without creating an asymmetry — where some slow down while others do not.

What This Means

The very fact that the letter appeared, and the scale of the debate that followed, is a signal. "Pacing the frontier" has ceased to be a marginal idea and has become a serious item on the AI community's agenda. As Mowshowitz emphasizes, the question is not whether this is good or bad in the abstract — but whether the industry is ready to make a considered decision before circumstances make it for them.