Recursive Self-Improvement, Straight Into Scrooge Musk’s Money Bin

Anthropic published When AI builds itself - Our progress towards recursive self-improvement, and its implications a couple of days ago. I was expecting a more in-depth account of how they, or should we call it “it,” built itself. Instead, it’s mostly stats on how much code is now being written by AI. I spent most of the piece thinking, “Duh… we already knew this.”

But I had missed the “and its implications” part of the subtitle. They buried the lead in the conclusion, and that’s where things got interesting.

If it were possible to effectively slow the development of this technology to give ourselves more time to deal with its immense implications, we think that would likely be a good thing. But if a slowdown simply lets the least cautious actors catch up technologically, it could leave everyone less safe. Without a global coordination mechanism, companies and governments will have to make difficult decisions about safety while under competitive and geopolitical pressures.

We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology. The Anthropic Institute will conduct research—in collaboration with many others—and take actions to help build the systems that a credible slowdown or pause would require. These systems would enable frontier AI developers to verify that others globally have actually stopped or slowed, and that a bad actor could not use the auspices of a coordinated slowdown to jump ahead in secret.

Who are these potentially bad actors? China? And is Anthropic itself truly a responsible, safety-conscious actor, genuinely worried about safety?

I then remembered something Sarah Friar, OpenAI’s CFO, said on the All-In Podcast a couple of days earlier: in 2026 and 2027, compute will be harder to get. Except for Elon.

Do they just want to slow down Elon?

Then I saw this: https://x.com/signulll/status/20630024328046718021

you’re telling me anthropic & google are paying spacex ~$26b a year for compute?!!

Ah! They’re just tired of filling Scrooge Musk’s money bin.

  1. The $26 billion a year figure checks out. Verification links: https://www.sec.gov/Archives/edgar/data/1181412/000162828026041150/spacexagreementfwp.htm, https://www.axios.com/2026/05/20/anthropic-spacex-compute, https://www.anthropic.com/news/higher-limits-spacex ↩