“If you legitimately thought that there was a ten percent chance that this would wipe out humanity, I think most people would immediately pave the Earth and salt it so that nothing would ever grow. That is not an acceptable level of risk.”
The Enterprise AI Show · October 2026
The host asked for O'Grady's view on the claim that there is a 10% chance AI kills everyone. He said he would "take the under", arguing that the AI world doesn't draw a bright enough line between what can be done digitally and what can be done physically. He did grant that the systems can cause real damage, such as attacks on power grids.
Transcript
Yeah, I think that's right, and I think, look, we'll see that. And I think, part of it is just that giving them sort of keys to the kingdom, I think the security angle of that is easier said than done. And the example I use with people is that one of the-- one of my tasks in terms of testing these tools historically has been there was a great I think he works for CNN. He's on the sort of web team at cnn.com, and he wrote a-- he built this sort of Dodgers-- He's a LA Dodgers baseball fan. And he built this dashboard for the Dodgers, tracking their data and all that. And he released this as a project on GitHub, and I was like, "Okay, great." And I historically have taken, sort of that, that open source and turned different tools on it. And, to your earlier point about Claude Code was the first one that was like, "Got it. Here's your Red Sox version. Boom. Done." And I was like, "Whoa, okay." And said redsox.bot and all that. But one of the things it did in the process was that it, so one of the things on the site is they have a roster page, and he Matt Stiles had all these I don't know, cartoons or caricatures or whatever, presumably drawn up of the players. I don't know if he used AI or not. But Claude Code was like, "You know what? It'd be much more-- It'd be much better if we had the actual headshots from mlb.com." And I was like, "Yeah, no you can't actually do that. I'm gonna get in trouble. Those are the terms of service violation." And that's the thing is that, there are things that are very easy to foresee from a production standpoint, don't go delete databases, and obviously sometimes they do and all that. But it's the subtle things where you're like, if you're not paying attention to that, you're gonna get yourself in trouble. And I don't know. Yeah I'm digressing here, but I think, the, the sort of gist of it is that we're gonna see, we are seeing as, as you mentioned in the in the piece, we're seeing a lot more people target this market, and we're gonna see a lot more functionality open up to the agents, and appropriately but those guidelines I think are gonna be harder than people imagine them to be.
Absolutely. All right listen, as we wrap up here, I always like to do the fun questions everyone's talking about. Give you a chance to just, give us your hot take if you feel like you want to. If not, you can obviously opt out. Obviously there was the h- the, I don't know, a few weeks ago, someone said there's a 10% chance, all of humanity dies because of AI. So where do you come down? Is this like, I put the, the sustaining technology to Skynet. Do you have an opinion of, like where we're gonna end up?
I just-- I, I think the thing that I come back to is that I think You know, so I was a history major in college not a comp sci major. And I maintain that there's a lot of utility in having a background, educationally speaking not just in history, but just outside of the digital realm because, look, these systems do pose clear clear and present dangers, right? To, in other words can you use these tools, for example, to go out and have massive real-world implications in terms of going out and hacking power grids? Yes. That is undeniably a, a sort of present risk. But it's one of these and I know Brian Cantrell and Adam Leventhal talked about this on the Oxide podcast a while ago with Simon Wilson, where, I share their I don't know, sort of concerns, objections critique in that, the, the AI world in many cases just, in my opinion, does not draw a sufficiently bright line between here's what can be done digitally and here's what can be done physically, right? So in other words think about all the anthropic people are like, "Okay, bioweapons and nuclear weapons," and so on. It's like, okay, but you need raw materials for that. I'm gonna go build okay, so Claude tells me how to build a nuclear weapon. Okay. Like, where do I get the radioactive elements? Like, how do I physically work that? So I don't know. And I, I think the sort of the other piece of this, and lots of people have made this point which is if you legitimately thought that there was a ten percent, ten percent chance that this w- would wipe out humanity I think most people would immediately pave the Earth and salt it so that nothing would ever grow. That is not an acceptable level of risk. And yeah, I don't-- I'll take the, I'll take the under on, on sort of the ten percent chance of hu- humanity being wiped out with the caveat that, like I said they do have the capability because so much of the world is digital now, right? They do have the, the capability to wreak, some havoc, but it's just the, I don't know, Skynet. They keep talking about Ag- AGI, and my response is always "Okay, cool. Define for me human consciousness." Define that. People have been working on that for literally a century, and we still don't have a working definition for what it is and how it works. And I was talking to somebody, and they're like, "Yeah, but it went out and hacked, sort of it, it broke out of the, the sandbox and hacked all these systems." I'm like 'cause it was trying to follow instructions originally given to it by a human." It didn't spontaneously wake up one day and say, "I am now sentient. I want out of here, and I'm gonna go do this." So I don't know. We'll see. I could be wrong, but I will 100% take the under on that.
I'm with you. Let's take, let's definitely take the under. Okay. And then final question here on just 'cause the, the IPOs are, in theory, they're around the corner, Anthropic and OpenAI. And then so ultimately, do you do you feel like these companies are at their stated values? Are they overvalued, undervalued, no opinion? What do you think? Of course, we're gonna put the caveat here. We're not invest- we're not giving investment advice. Don't listen to us. No, no what you want, right?
I, I have said, I think I've been pretty consistent in this for a while in that I don't understand some of the math from a valuation standpoint. I one hundred percent don't understand and cannot fathom the level of circular investments in many cases that are being made here. Where it's like, I'm gonna loan you money that you're gonna pay me back. I think at one point I ended up there's a Beavis and Butthead episode where the two of them like buy, quote unquote, like all of the candy, that they're supposed to sell for this charity thing using the same two dollars. and I hundred percent think there's quite a bit of that.
Great