Can We Still Build AI Safely? The White House Thinks So | MOONSHOTS #298

So let's open with the biggest story of the week, and I wanna take it in four parts.

Uh, first, the backstory on this rollercoaster.

You'll remember two weeks ago, Dario, Sam, Elon, and Demis all called on the industry to pace the frontier, to slow down.

Then last week, President Trump told the reporters he wanted to leave it exactly where it was, that the DOJ acting as guardrails, right? We have liability laws.

We have protections.

We don't need to slow it down.

This Tuesday, Trump brought the entire AI stack into one room.

Nearly 20 tech leaders joined the president and Speaker Mike Johnson for lunch in the East Room.

Let's, uh, show the quick slide here.

What I found fascinating about this, uh, this image of all the leaders together at the White House is the seating arrangement, right? There is Trump w- with, uh, Jensen to one side and Elon to the other, and then in the far left, [laughs] you see Dario.

I think he's been, uh, sent to the outer pastures.

Uh, pretty interesting.

All right.

Well, you know, uh, Elon, Zuck, Jensen, Dario, Sundar, Satya, Lisa Su, and OpenAI President Greg Brockman were there.

Sam was in San Francisco for OpenAI's DevDay, uh, which we'll get to in a little bit.

So what happened? The White House signed the Superintelligence Accord, right, in four layers.

Let me read them out.

So layer number one, internal controls that monitor capabilities and alignment during training and deployment in cyber, bio, and chemical so models do not hack or access technical systems in unintended ways.

Layer number two, an internal team to make sure the controls are actually working.

Three, an external auditor or evaluator to independently assess them, and four, an independent committee of each company's board of directors to oversee it and make sure the problems get fixed.

Let me show a quick video here of Donald Trump speaking about this.

It really is a, a form of protection.

They understand that they have to self-police, and it's very hard for somebody to come out and say, "Get into these mo-..." These models are very complex.

And again, I believe they're gonna be used for the good, and when they're not, we're gonna be able to nab them.

[laughs]