The Signal.
The signal, not the noise.
Today
The Signal.
Desks
The paper
Ideas · Essay

AI’s slowdown may not slow much at all. The real effect could be an acceleration.

Anthropic, OpenAI, and xAI are backing new safety measures, but the central question is whether they actually delay progress or simply repackage it. Tyler Cowen says the public may be reading the move backward.
Monday, September 14, 2026

Anthropic CEO and co-founder Dario Amodei set off the weekend debate with a post titled 'We Must Pace the Frontier.' He said, 'We must slow the pace at which we improve the capabilities of AI models,' and proposed other safety measures that Anthropic is beginning to implement.

Sam Altman’s OpenAI and Elon Musk’s xAI have endorsed the proposals. They include embedding third-party monitors in the companies and placing stricter restraints on exports to China. The most discussed piece is the so-called 'slowdown,' which could either help or hurt AI progress depending on how it is carried out.

Tyler Cowen, who says he is a member of Anthropic’s economic advisory committee, argues that much of the public has misunderstood what is being proposed. In his view, this is not an actual slowdown. It is a temporary pause that could be followed by a later acceleration.

Cowen’s reading comes after a public warning from Jacob Coxon, an AI researcher who resigned from Anthropic and said the company was not doing enough to make sure AI functions in keeping with its designers’ intentions, a problem known as alignment. Coxon accused Anthropic and OpenAI of 'gambling with our lives.' His post has since drawn more than 169 million views.

Evan Hubinger, who still works on alignment at Anthropic, put the probability of the world ending because of AI at more than 10 percent within the next decade. Senator Bernie Sanders announced he would seek a slowdown of further AI development and a ban on superintelligence.

Cowen says he is an avowed AI optimist and a skeptic of most proposals to forcibly restrict progress, but he is open to the changes major AI labs are now endorsing. His point is narrower: what is being sold as restraint may not function that way in practice.

That distinction matters. If the measures are mostly cosmetic, then the public gets the language of caution without the substance of delay. If they genuinely change incentives inside the labs, then they may improve safety, but at the cost of progress that markets, firms, and users are already racing to capture.

Follow the incentive, not the press release. The record is public: the companies say they are slowing the pace, but the same companies are also building the next stage of the race. What is true does not need an adjective.

More from Ideas