Major AI companies have tentatively agreed to pace the frontier — slowing down the development and release of their most capable models. The push was catalyzed by Anthropic CEO Dario Amodei's long essay published this weekend, titled We Must Pace the Frontier, which argued for deliberate deceleration and coordination between governments and AI labs.
The proposal has drawn broad support from AI executives. But critics — including some antitrust scholars and competing startups — are raising pointed questions about whether a coordinated slowdown among dominant market players constitutes a safety measure or an illegal cartel.
The core tension: when the companies controlling the vast majority of AI compute and talent agree to slow down together, the stated goal is safety, but the structural effect is to raise barriers to entry for anyone who doesn't agree to the same terms. As one critic put it, a speed limit only works when everyone agrees to it — and the people setting it happen to be the ones who are already ahead.
The debate comes at a pivotal moment. Frontier model development has accelerated to the point where new capabilities emerge faster than regulators can assess them. Yet the proposed self-regulation mechanism has no enforcement teeth, no independent oversight, and no mechanism for smaller players to participate as equals.
Supporters argue that waiting for governments to impose rules would be slower and clumsier, and that industry coordination — however imperfect — is better than uncontrolled acceleration. Detractors counter that voluntary commitments from companies with fiduciary duties to maximize shareholder value are, at best, aspirational.
