OpenAI made hesitation look reckless, then helped pull companies and governments into an infrastructure race. Now Altman, Amodei, and Hassabis want ways to slow the frontier. The skeptics warned them years ago.
Sponsor
Create your digital twin
Your Sentience is a personal AI that understands your values, knows what you care about, and acts proactively on your behalf.The AI pitch used to be simple: move now or lose the future.
Sam Altman said humanity had crossed the event horizon of a “gentle singularity”. He forecast agents doing real cognitive work, producing scientific insights, and entering the physical world as robots. Intelligence and energy were heading toward abundance.
Then OpenAI put a price on that confidence. Stargate promised $500 billion of AI infrastructure, along with enormous economic gains and a path to broadly beneficial AGI.
The message reached far beyond OpenAI. Anthropic urged the United States to capture the economic benefits of powerful AI. Google DeepMind presented Demis Hassabis’s universal assistant as a boost for productivity, daily life, and science.
That message pushed investors, corporate boards, and governments toward chips, power contracts, and data centers. Companies that waited risked becoming irrelevant.
The money moved.
Now several of the people who sold that urgency want a brake.
The reversal
Bloomberg reported that Altman told OpenAI employees the company could pace cutting-edge development if rival labs moved with it.
Axios reported that Hassabis proposed a government-supervised body with advance access to frontier models and the ability to coordinate a slowdown when danger rises.
Dario Amodei has made the strongest case for slowing capability development so safety work can catch up. Axios described him as one of the industry’s loudest voices on catastrophic risk, so his concern is hardly new. His remedy is. Anthropic built its identity around keeping safety inside the race. Amodei now argues that the race itself may need pacing.
No one in this group is proposing a permanent shutdown. Their plans differ, and major leaders still reject the premise. Nvidia’s Jensen Huang and Meta’s Mark Zuckerberg have pushed back against the new alarm.
The reversal remains stark. Frontier labs spent years treating speed as an economic necessity. Several now want coordinated delay treated as responsible policy.
The warnings were already public
The risks did not arrive with the slowdown proposals.
Jan Leike left OpenAI in 2024 after saying safety culture had taken a back seat to “shiny products.” OpenAI dissolved his Superalignment team. The dispute was public, and the company kept shipping.
Five months later, Miles Brundage left OpenAI. Fortune reported his warning that neither OpenAI, other frontier labs, nor the world was ready for highly capable systems.
Other critics focused on the damage already visible. The 2021 “Stochastic Parrots” paper identified the environmental cost, bias, and concentration of power tied to ever-larger language models. Those harms were already present in the system being built.
Market skeptics questioned the bill. A 2024 Sequoia analysis calculated that the industry would need roughly $600 billion in annual revenue to justify its infrastructure spending. The arithmetic exposed the gap between the products people were buying and the capacity companies were building.
These critics disagreed about the main danger. Some feared loss of control. Others focused on labor, surveillance, copyright, energy use, and corporate power. The market critics doubted the economics. All of them saw pressure building while the labs kept telling the world to accelerate.
The brake comes after the moat
The timing matters because the leaders keep what the sprint bought them.
Years of urgency pulled capital toward a small group of companies. That capital bought compute, data centers, talent, customers, and influence over government policy. The race made restraint harder, then rewarded the companies best equipped to survive it.
A coordinated slowdown would begin from today’s leaderboard. OpenAI, Anthropic, and Google DeepMind would keep their infrastructure and market access. Smaller competitors would face safety rules written after the incumbents had already paid the largest entry costs.
The safety case may be sincere. OpenAI recently delayed a release over security concerns, and new evidence can change a responsible person’s mind.
Sincerity does not settle who benefits. Expensive licensing, private evaluations, and compute thresholds can reduce risk while protecting the companies already in front. Tech Policy Press made the conflict explicit: allowing the winners to shape the slowdown can preserve their advantage.
That is the issue the industry’s safety debate keeps avoiding. The leaders sold urgency when urgency raised money. They discovered the need for coordination after the spending race had built their moats.
What proof would look like
Companies serious about slowing down can publish the capability thresholds that trigger a delay. They can give outside evaluators access and the right to publish bad results. They can disclose when a release was paused and what changed before it shipped. They can accept rules written by public institutions with workers, researchers, smaller firms, and open-source developers at the table.
Most important, they can support rules that do not freeze the current leaderboard.
Without those commitments, “safety” is also an excellent incumbent strategy.
The bill is already here
Changing your mind about a dangerous technology can be wisdom. This change did not happen in private.
Frontier leaders broadcast certainty. They made delay sound ruinous. Their forecasts helped turn AI investment into a corporate and national obligation. The public was invited to finance the race before it was invited to debate the brake.
Now the people who sold the urgency want authority over the slowdown. Public institutions should write the rules. The companies whose spending race created the problem should not control them.