Leaders and researchers building artificial intelligence are beginning to ask whether the race is moving faster than our ability to control it.
Published on September 19, 2026 · by Alex · Artificial intelligence · AI safety · Anthropic · Nvidia
Something unusual is happening inside artificial intelligence: some of the people competing to build the world's most advanced systems are saying that perhaps development should slow down. The uncomfortable question is what they are seeing that changed the conversation.
This article is the written version of the Saturday analysis.
Watch the video on TikTok · Follow the channel
The debate gained visibility after Jacob Coxon, a researcher at Anthropic, publicly resigned and warned about the risks of increasingly autonomous systems. His departure was described in press coverage as a sign that the capabilities race may be moving faster than safety controls. The Washington Post reported on his resignation and concerns.
Days later, Dario Amodei, Anthropic's chief executive, published “We Must Pace the Frontier”. He did not call for abandoning research. He argued for pacing frontier-model development, stronger evaluation and time for safety measures, including work by external evaluators.
The conversation also received partial support from figures such as Sam Altman and Elon Musk, according to public coverage of those statements. That does not amount to a common plan or a formal agreement to stop development.
This is unusual because these companies compete for talent, customers, capital and the lead in better systems. Some technology leaders have supported parts of these concerns, but that is not the same as a formal agreement to stop the race. The difficulty appears when concern must become an operational decision: if one company slows down while others continue, who bears the cost?
There is also a geopolitical question. What happens if the United States slows innovation while China continues developing its systems? Can everyone promise safety when nobody wants to lose the race? Everyone may agree that risks exist, but they may not agree on who should slow down first.
On one side, Amodei and other leaders have argued that safety must catch up with the frontier. On the other, Nvidia chief executive Jensen Huang has publicly argued that agentic systems are already increasing productivity and that more computing capacity can produce more useful output. In an earnings call, Huang described improving agents and reasoning as a trend already under way. Nvidia's earnings transcript records that position.
The difference is not simply between optimists and pessimists. One view says we are approaching capabilities we do not yet know how to control fully; the other says the answer is not slowing development but building better systems and moving forward with more capacity.
For the first time, this debate is taking place inside the companies building these systems, not only between outside critics and Silicon Valley. That does not prove AI will collapse. It does require watching specific variables: release speed, safety incidents, regulation, computing costs, data-center dependence and who pays if a product or infrastructure project must be delayed.
A public statement must also be separated from an operational change. A company can ask for caution while hiring researchers, buying chips and expanding capacity. The important signal will be whether verifiable limits, independent evaluations and safety budgets appear and survive competitive pressure.
Perhaps everyone is waiting for someone to hit the brakes, but nobody wants to be first while others accelerate. Some statements may also seek to influence regulation, component prices or market expectations. That is a hypothesis, not a demonstrated fact.
So the concrete question to follow over the next few months is who actually starts slowing artificial intelligence and who merely says that others should do it. The answer will show whether the industry is building controls that can keep pace or whether the race remains ahead of them.
Sources: Dario Amodei, “We Must Pace the Frontier”; Nvidia, fiscal 2027 second-quarter earnings transcript; The Washington Post, coverage of Jacob Coxon; Inside Wall Street TikTok video. References to coordinated slowdown, China and a pricing strategy are questions and scenarios, not conclusions.
He wrote publicly that he was concerned AI companies were prioritizing the race for more capable systems without yet having enough controls for increasingly autonomous systems.
He called for pacing frontier development and giving evaluations and safety measures more time, including work by external evaluators.
His public position is different: he has emphasized that agentic systems and computing demand are growing, and that the answer should be building more capable and productive systems rather than simply stopping.
Model-release speed, computing costs, safety incidents, regulation and who pays if a product or infrastructure project has to be delayed.
