Skip to content

Ivana Nikolić Hughes: AI Meets Nuclear Weapons (Transcript)

In this interview, Ivana Nikolić Hughes, president of the Nuclear Age Peace Foundation and senior lecturer at Columbia University, joins Glenn Diesen to discuss the alarming dangers of combining artificial intelligence with nuclear weapons. They explore nuclear winter, Cold War close calls, the collapse of arms control, the new nuclear arms race, and what a path to a world free of nuclear weapons could look like. This interview was premiered on October 8, 2026. Read the full transcript of this interview below:

Introduction

GLENN DIESEN (00:00:00 – 00:00:14): Welcome back, everyone. Today we are joined by Ivana Hughes, the president of the Nuclear Age Peace Foundation and also a senior lecturer at Columbia University. Thank you for coming on the program.

IVANA NIKOLIĆ HUGHES (00:00:15 – 00:00:18): Thank you so much for having me, Glenn. It’s great to be with you.

When Nuclear Weapons Meet Artificial Intelligence

GLENN DIESEN (00:00:19 – 00:01:20): So we were talking earlier in your area of work, that is with nuclear weapons, it seems Yeah, sadly, it’s not becoming less relevant, although people seem to focus less on it than they did during the Cold War. And all this, I have a lot of questions about the development because we see a lot of changes and challenges we didn’t even have during the Cold War, as nuclear weapons isn’t becoming less relevant. Indeed, they’re becoming more unpredictable and largely more relevant. And sadly, it’s more likely that they will be used, it seems.

But my first question to you though was about an article you recently wrote. You wrote about essentially what happens when nuclear weapons meet artificial intelligence. Again, 2 of the things which are often cited as possibly putting an end to humanity. I was wondering if you could flesh out that argument or why you see this as a very dangerous combination.

IVANA NIKOLIĆ HUGHES (00:01:21 – 00:11:11): Absolutely. Thank you so much for that intro. And I just want to say that I personally became interested in nuclear weapons about 15 years ago. Actually, I was already a faculty member and teaching at Columbia, and sort of my education and research and so on had been in other areas. And at the time, it seemed like a really irrelevant topic almost. The remnant of the Cold War. Sure, these weapons were still around, but nothing had happened in 60-plus years. And now we see them becoming more and more relevant, not year by year, quite literally day by day. And so there are a number of recent developments that are incredibly concerning, and I think we’ll get to some of them.

But with this, the talk of AI And of course, AI has now these large language models at the kind of hands of an everyday person have been in our lives for the last few years. And I think ever since that, people in the nuclear disarmament community have really been concerned about this notion that AI could in fact become involved, engaged in either aspects or sort of fully in nuclear command and control systems and recognizing from the very beginning that AI, for example, could hallucinate, right? That they could, it could sort of make erroneous decisions. And so people have been concerned ever since that if in fact governments decide to engage AI in this way, that we could end up inadvertently in some kind of nuclear weapons use.

And just to say very broadly, war games that are conducted regularly in Washington, D.C. suggest that any use of a nuclear weapon will actually spiral into a full-blown nuclear war. So that, I’ll just leave that aside. So this notion that AI, that we could deliberately decide to involve it in nuclear-related decisions and that it could do something inadvertently.

Then there was a King’s College study about a year ago or in the last year, which suggested where they ran 21 scenarios and asked 3 different models. I don’t remember exactly what they were. One was from OpenAI and the other 2 from some of the other top companies. And those AI were in these, asked to sort of make decisions and a series of decisions in these scenarios. And in 95% of the games scenarios, they decided to use a tactical nuclear weapon. And in 76% of them, they decided to do the strategic nuclear signaling to kind of proceed with strategic. And the signaling didn’t seem to sort of calm the situation down, it seemed to just sort of escalate it further.

So again, the idea that we would decide to involve AI in these kinds of decisions is in and of itself incredibly dangerous, risky. I mean, just in so many ways catastrophic. And we should talk a little bit about what that means. What does actually nuclear war mean?

But to come back to the last I would say a couple of months since the revelations that the OpenAI model had sort of hacked into this Hugging Face. That was sort of the first incident. But since then we’ve had a number of incidents. It’s almost hard, now they’re not top news, front page news. They’re sort of in the back, in some back article. A recent one I saw, I think in the last week was from Australia where the AI models had hacked into the government sites in Australia, their kind of Medicare system, obviously going in and having access to data, private health data and so on.

But really demonstrating that these models are now doing something beyond just we tell them, go ahead, do X, Y and Z, and they either make a bad decision or they erroneously or hallucinatingly do something bad. This is now them going around the constraints that humans had imposed and collaborating with one another and hacking into these systems, raising really this question of, are they going to hack into a nuclear command and control system? Again, whether it’s an intentional or non-intentional, and what are we really doing about preventing that from happening?

And combined with some of the inherent vulnerabilities in these nuclear weapons, the way that these arsenals exist, the way that they’re meant to be delivered, for example, intercontinental ballistic missiles in the United States, which it’s well known where those warheads are, where those missiles are.