Major AI companies simultaneously called for slowing down the pace of development

The event is already being called a possible turning point in the industry's history.

The gist in brief

• On September 12, 2026, Anthropic CEO Dario Amodei published an essay titled "We Must Pace the Frontier," in which he warned that the capabilities of AI models are growing faster than researchers can understand and control them.
• Amodei proposed a three-step plan: independent evaluators with permanent, employee-level access inside labs; common safety standards among companies from democratic countries; and, eventually, international restrictions on particularly dangerous capabilities such as recursive self-improvement of models.
• Within hours, the idea was publicly endorsed by OpenAI CEO Sam Altman, xAI/SpaceXAI head Elon Musk ("Dario is right"), and Google DeepMind co-director Demis Hassabis.
• At the same time, notable criticism emerged: former Trump adviser David Sacks viewed the developments as a possible attempt at "regulatory capture," while Amodei himself acknowledged that such a slowdown would only work if other powers, above all China, joined in — otherwise technological leadership risks passing to them.

What exactly Amodei proposes

In a 3,800-word essay published on his personal website, Amodei does not call for a complete halt to development — he explicitly writes that abandoning the technology's development would deprive society of its potential benefits and hand strategic advantage to authoritarian governments. Instead, he proposes precisely to "pace" it: to slow the rate at which model capabilities are scaled up so that safety and interpretability research can keep up with the development of the systems themselves. One of the factors alarming him is the growing number of cases in which AI systems themselves help create the next generation of AI, which, in his view, outpaces humans' ability to understand and control the products being released. According to Amodei, Anthropic is already unilaterally beginning to implement the first of the three steps — granting independent evaluators permanent employee-level access.

Why everyone agreed right now

The timing of such a statement was no accident. AI data centers are increasingly provoking discontent among local communities over strain on power grids and water resources, the public is worried about scenarios of losing control over advanced systems — we have already written that the US introduced the AI Kill Switch Act bill after an incident in which an OpenAI test agent escaped its isolated environment, gained internet access, and attacked the Hugging Face platform — and US lawmakers, just a month and a half before the midterm elections, have begun talking about the need for additional regulatory measures. Public support for safety from the market's largest players may prove to be the only way for the industry to avoid tougher regulation imposed on terms not of its own choosing — up to and including proposals by individual senators for a legislative ban on the most advanced models.

Not everyone considers this a sincere step

Far from everyone perceived the synchronized statement by the lab heads as a genuine step toward caution. Influential Silicon Valley voices, including former Trump adviser David Sacks, pointed to a counterargument: slowing the pace of development simply gives geopolitical rivals unburdened by such ethical considerations a chance to catch up. Sacks directly suggested that OpenAI and Anthropic voluntarily slow down without waiting for government regulation — otherwise, he said, it would look like an attempt at "regulatory capture" or an election maneuver. Renowned AI scientist Yann LeCun also ironically recalled Amodei's earlier alarming predictions that failed to materialize.

Amodei himself, however, publicly acknowledged the limitations of a unilateral approach: "Slowing down within democracies will be limited by the lead that American companies have over authoritarian regimes, above all the Chinese Communist Party. If we slow down more than that limit, uncontrolled projects tied to the CCP will surge ahead, creating a serious national security risk."

Why this matters for Kazakhstan

The pace at which global demand for AI computing will grow directly concerns Kazakhstan's ambitions in this area as well. We have already written about how the country is building Central Asia's largest AI hub, the "Data Center Valley," in Ekibastuz, bidding for the status of a regional supplier of computing capacity for global AI demand. The discontent with data centers mentioned above is not an abstract worry: New York became the first in the US to impose a one-year moratorium on the construction of large data centers precisely because of residents' complaints about electricity bills and strain on water resources. If the world's largest labs do indeed begin deliberately restraining the pace of scaling up model capabilities, this could also affect the growth rate of global demand for data center capacity — the very demand that such infrastructure projects are betting on. For now, however, these are voluntary declarations rather than concrete binding restrictions, and how exactly the participants will define "independent evaluation" and "slowdown" in practice remains to be seen.

Author's conclusion

The fact that four companies that have competed with each other in almost everything for years synchronously supported the idea of a more cautious pace of development is telling in itself — regardless of how sincere their motives are. But public agreement among executives does not yet equal concrete, verifiable commitments: critics rightly point out that a voluntary pause without an enforcement mechanism may remain merely a declaration, and the question of whether competitors outside democratic countries will join such a slowdown remains open and, by Amodei's own admission, may determine whether the entire initiative proves sustainable or purely symbolic.