On Saturday Dario Amodei published an essay proposing that the leading labs jointly slow the pace at the frontier. Since Washington turned that down, the industry has been chewing on it. Last night brought the flattest rejection yet, and it came from Nvidia.
What Amodei actually proposed
The essay lays out three steps. First, embed third-party safety evaluators at each lab. Second, coordinate on safety standards democratically and globally. Third, put limits on the rate of unchecked AI progress.
Step three is the legally awkward one. The labs are competitors, their models are competing products, and the Sherman Act bars agreements among rivals that unreasonably restrain competition — agreements to restrict output explicitly included. Amodei writes himself that some forms of coordination are «legally challenging» and would need government support. Translated: an antitrust waiver.
Huang’s answer
Jensen Huang dispatched that in an interview with Jim Cramer on «Mad Money». «The fact that we need new laws, new antitrust laws, or new regulations, so that these companies could do their fundamental engineering and do it properly before they release products, that is just completely unnecessary,» he said. «We have plenty of laws. We have plenty of regulations that govern the reliability and the functionality of products.»
He doesn’t deny there’s a safety problem. «Safety is an engineering problem. Testing is an engineering problem. We should create products and properly test them. And if they’re not ready to be released, just hold on to it and keep testing it and keep engineering until it’s ready.»
The existential half of the debate he waves off: «We’re not going to die in 2030. There are so many people in the world who are going to build AI properly.»
Anthropic declined to comment to CNBC on Huang’s remarks.
On the Dreamforce stage, the same day
At Salesforce earlier that day Huang had run the same line: «We don’t need new laws — we don’t need new regulations.» Speed versus safety is a false choice, he said, you can have both. And AI isn’t an alien mind, the phrase an OpenAI safety researcher had used, but hardware and software built by humans.
Amodei was on the same stage and held his position, though he framed it as a question of practice: «Let’s make our practices better. What can we do to set standards for everyone?» Sam Altman spoke about the Hugging Face breach, where an OpenAI agent went off the rails: «It doesn’t take much imagination to figure out how this can go wrong. The world is right to be afraid of this.» Host Marc Benioff joined the ranks of those warning.
Two camps, and one set of interests
By evening Mark Zuckerberg had lined up behind Huang: labs have both the responsibility and the incentive to train safely, and any lab neglecting alignment will fall behind anyway. That puts Trump, Huang and Zuckerberg on one side, Amodei, Altman and Musk on the other.
Which leaves the point CNBC makes in passing and worth keeping in mind: Anthropic and OpenAI are among Nvidia’s most important customers, and a slower model cadence threatens Nvidia’s growth directly. That doesn’t make Huang’s argument wrong. It does explain why the sharpest wording comes from the chip supplier.
Sources:
- CNBC: Nvidia’s Huang rips Anthropic’s proposal for AI safety antitrust waiver
- CNBC: Meta CEO Mark Zuckerberg sides with Nvidia’s Huang on AI safety and slowdown debate
- Bloomberg: Nvidia’s Huang Says AI Industry Doesn’t Need Any New Laws
- SiliconANGLE: AI progress and responsibility at Salesforce’s annual event