Five AI Leaders Agree Development Should Slow Down

Five of the most prominent leaders in AI agree that development should slow down before control is lost.

That agreement is the throughline of a feature published Sept. 16, 2026, by The Verge titled 'A brief history of AI executives calling for regulation'. The feature names OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Google DeepMind cofounder Demis Hassabis, Microsoft CEO Satya Nadella and X CEO Elon Musk as sharing that position.

The record it traces starts in January 2015, when Musk joined Stephen Hawking in signing an open letter asking for responsible AI research. No specific policy tool was named, only the request itself. The signatures came early, before large-scale commercial use of the advanced systems now called frontier AI.

The next dated point comes in July 2017. Musk told a gathering of US governors that AI needed to be regulated right away. By that same month, he had invested $38 million in OpenAI, according to the same account. The call for limits and the investment existed side by side.

The broader context here is why that lineup stands out. Altman, Amodei, Hassabis, Nadella and Musk do not run the same kind of operation. They span research labs, platforms and infrastructure. Agreement across that range is unusual in enterprise technology, where vendors more often say existing law and internal review are enough. It fits a familiar pattern from earlier platform shifts. Builders ask for rules while they are still building. Regulators, customers and standards bodies then must decide what counts as proof of safety, who can check it, and what happens when tests, deployment limits and incident reporting add real cost and delay.

In my view, the useful question is not whether executives mean it when they call for regulation. Incentives are mixed by definition. A lab that wants guardrails may also want predictability, clear liability rules, and requirements that favor well-funded teams with mature testing and governance tools. Those motives can sit alongside real concern about losing control. For people building systems, the practical response is to treat executive statements as one input and focus on workable detail.

Looking at what this means for working technologists, the work sits below open letters and meetings with governors. Versioned model cards, which are standard records describing what a model does and its limits, reproducible tests, staged rollouts, logging and rollback plans, access controls around model weights and training computers, and outside review are where a call to slow down becomes real or stays talk. If regulation follows the path sketched in this history, those items become compliance material, not just good engineering.

There is a longer arc worth keeping in mind. I watched my own children grow up with changing defaults, from desktop software to always-on phones to cloud services that update without asking. Each shift first looked hard to govern. Each later gained norms, tools and laws that made it more routine and more dependable. AI may follow a similar path, with early friction and better building blocks later.

For now, the facts are narrow but consistent. The thread runs from a 2015 letter on responsible research, through a 2017 call for immediate regulation, to a 2026 summary in which five prominent leaders agree development should slow before control is lost.