Sam Altman told OpenAI staff this week that the company is considering slowing down development of cutting-edge artificial intelligence, and that he hopes other companies will do the same. The reported condition attached to it is the part worth reading twice: if the government confirms that coordinating on pace is not an antitrust problem.
The week around that remark was loud. An Anthropic researcher resigned publicly over concerns that AI labs are gambling with our lives. OpenAI’s chief scientist called for a research slowdown. Anthropic’s Dario Amodei published a long post committing his company to new safety steps including third-party evaluators and calling on the industry to downshift, Altman pledged to adopt the independent-evaluator suggestion, and Elon Musk wrote that Dario is right.
SiliconANGLE’s Robert Hof, who has watched this industry for a long time, put the obvious objection in one line: we will see whether any of this existential worry changes anything, and he will believe it when he sees it. In the same week, new models, agents and apps shipped from Meta, OpenAI twice, DeepSeek and others.
That is not hypocrisy. It is the structure of the problem, and the antitrust clause gives it away completely.
Consider a street of coffee shops. Every operator on it privately agrees that opening at five in the morning is bad for everyone. The staff hate it, the early trade is thin, the margin on those two hours barely covers the wage bill, and it exists solely because one shop did it and the rest had to follow. Get them in a room and they will all say the same thing.
Now try to fix it. Any single owner who moves to a seven o’clock opening tomorrow hands two hours of trade to the shop next door, permanently, because customers form routines and do not come back. So nobody moves. The only thing that would work is all of them agreeing together, and all of them agreeing together on trading hours is exactly the thing competition law exists to prevent.
This is the shape of every coordination problem in commerce, and it has almost nothing to do with whether the participants are sincere. Sincerity is cheap and abundant here. What is scarce is a mechanism by which the first mover does not simply lose. For two years the industry’s answer to every call for restraint was that slowing down was impossible because of China, which was a competitive argument wearing a geopolitical coat. The argument has not gone away. It has been joined by a new one, which is that even domestic coordination may be illegal.
So what the labs are actually asking for is not permission to be careful. They are asking for cover, in the specific sense of an external authority that makes restraint mandatory for everyone at once and therefore costless for each. Regulation, in this reading, is not something being resisted. It is being requested, because it is the only instrument that solves a prisoner’s dilemma the participants cannot solve themselves.
There is a second constraint doing quiet work underneath all of this, and it might matter more than the ethics. Compute is bottlenecked by power supply chains, and community opposition to new data centres is rising. Some deceleration may arrive whether anyone chooses it or not.
Hof’s scepticism is the right posture until behaviour changes. The week’s shipping schedule is the evidence that matters, and it did not change.
Every machine on the street is still running. They are all watching each other’s doorway.
Source: SiliconANGLE, “AI’s existential crisis explodes but AI companies plunge ahead anyway”
