Anthropic CEO Dario Amodei now says AI companies need to deliberately slow down frontier AI development.

And this time, he’s proposing an actual plan for doing it.

Amodei says AI capabilities have recently been advancing “drastically faster,” particularly when it comes to AI systems helping build the next generation of AI.

I'm pretty sure tho, that he wants government regulations so that smaller AI businesses get throttled, so it becomes almost impossible for them to catch up.

I'm not against safety, I'm just not that into transparent tactics to gain an unfair advantage.

He argues that labs need to slow the rate at which those capabilities improve while using the extra time to strengthen safety.

Here’s the real story:

Anthropic wants independent safety evaluators embedded inside frontier AI companies, with access similar to internal risk teams. Anthropic says it will voluntarily do this, and Sam Altman says OpenAI plans to follow.

But, I'm willing to bet it's 25% safety and 75% "danger marketing," an approach Amodei uses a lot to push his regulatory regime strategy.

He wants government regulations so that smaller AI businesses get throttled, so it becomes almost impossible for. them to catch up.

This is often referred to "pulling the ladder up," meaning that now Open AI and Anthropic are waaay ahead, the pull up a figurative ladder to limit the ability to climb.

Amodei also wants leading AI companies to coordinate on common safety standards and limits on unchecked development, potentially with the U.S. government enabling those discussions.

Eventually, he wants some level of global coordination, including limited agreements with China around especially dangerous AI applications.

But there’s another side to this.

Critics argue rules like these could disproportionately benefit the biggest frontier labs by creating regulatory barriers that smaller competitors can’t easily absorb.

Amodei’s plan also calls for restricting advanced chips and semiconductor technology to China, which he argues could widen America’s AI lead over the next three to five years.

So the question isn’t simply whether AI should slow down.

It’s who gets to decide the speed limit, and who benefits once it exists.