Emerging Situation: Anthropic's Global Pause, Recursive Self-Improvement Arrives, and AI Personhood Arrives | EP #263

**Peter Diamandis** (0:00)
Anthropic just published what I think is probably one of the most important documents an AI company has ever released. More than 80% of the code merged into their code base is written by Claude.

**Alexander Wissner-Gross** (0:10)
The AI is building itself. The AI is recursively self-improving.

**Dave Blundin** (0:15)
Think back on the turning points of humanity, World War II, Cuban Missile Crisis, nuclear weapons. This is so much bigger.

**Peter Diamandis** (0:21)
Argentina invites AI to free itself, positioning Argentina as the global haven for AI companies, new regulation, new legal personhood for machines and open for business.

**Salim Ismail** (0:32)
I think it's a very bold move. Whoever doesn't gets first-never advantage.

**Alexander Wissner-Gross** (0:36)
This is obviously a huge step forward for AI personhood. There will finally be a legal home for you on this planet.

**Peter Diamandis** (0:44)
This is an inflection point. This is a moment in time that the world just changed.

**Dave Blundin** (0:53)
Now, that's the Moonshot, ladies and gentlemen.

**Peter Diamandis** (0:58)
Welcome everybody to Moonshots. Today is a special episode. There were three stunning stories we want to report on.
And we pulled the mates together. Gentlemen, good morning. Salim, where on the planet are you? You're just a probability function.

**Salim Ismail** (1:10)
This is actually a real background behind me. I'm at Dallas-Fort Worth Airport. I came from the McKinsey's AI Summit in Monterrey, Mexico yesterday.
I'm on my way to Vegas tomorrow. But like, look at this, how apropos.

**Peter Diamandis** (1:21)
That's amazing. And Dave, how about you, Bell?

**Dave Blundin** (1:25)
Back at Stanford, last night was AI demo day, shark tank style, 10 back-to-back. Incredible, actually, what the students are able to build. The winner, well, one of the winners had started working on their product at 11 o'clock last night.
They made the mistake of mentioning that in their presentation, which dinged their score a little bit, but it was incredible what they got done. It's actually called an archive. It's like a replacement for archive where agents can read the papers instead of people. And they got it cranked out every night. It's unbelievably impressive. So just amazing compared to a year ago. Yeah. Wow. Well, it ties into today's stories too.

**Peter Diamandis** (2:00)
Amazing. And Alex, you and I look like we're in a normal haunts.

**Alexander Wissner-Gross** (2:05)
I'm in cyberspace.

**Peter Diamandis** (2:07)
Okay. Good for you. All right.
Let's get started. The news of Anthropic yesterday was stunning. Let me set the stage for we're going to be discussing for the next 30 minutes unpacking. Anthropic just published what I think is probably one of the most important documents an AI company has ever released. And it's the reason we're recording this unscheduled Moonshots podcast today. The paper is called When AI Builds Itself. And it's from the Anthropic Institute, written by Marina Favreau and Jack Clark. Here's what they're saying. AI is already accelerating development of AI itself, not theoretically, not in the lab someplace. It's happening right now inside Anthropic. More than 80% of the code merged into their code base is written by Claude. Their engineers are shipping 8x as much code per quarter as they were, you know, just a year back. Claude Opus 4.6 can now handle tasks that take a skilled human 12 hours. A year ago, it could handle 4 minutes. If the trend holds, Anthropic is saying that by the end of 2027, Claude will be able to do week-long tasks on its own. And here's the stunning news. After you put all this data together, after the trajectory charts, a company that's about to IPO for a trillion dollars with 640% user growth is causing a stir by calling for a temporary pause on global frontier AI.
Marina Favreau and Jack Clark wrote, quote, We believe it would be good for the world to have the option to slow or temporally pause frontier AI development to enable societal structures and alignment research to keep up with the advancements of this technology.
Gents, I read that, Alex, you sent me that tweet and I was like, okay, we need to talk about this. Alex, let's go to you first.

**Alexander Wissner-Gross** (3:58)
You know, I define the singularity sometimes tongue-in-cheek as all sci-fi tropes happening everywhere all at once. Well, there's a universe in which we get UBI or UBE at the same time as the federal government getting golden shares in the frontier labs at the same time that we get recursive self-improvement at the same time that we get an AI pause all happening at once or at least an AI slowdown. It would require them all actually happening at once, I think, for the latter part of that announcement, the notion of Anthropic calling for a global coordinated slowdown in recursive self-improvement research. But it's not impossible. It's not necessarily desirable, but it's not impossible.