Special Investigations

Press Freedom Defense Fund

Trump’s AI-Powered World Wars

Nick Turse and Hooman Majd discuss war on Iran and other U.S. conflicts, and Sam Biddle breaks down how AI is being used.

In the last few days, President Donald Trump has said that the U.S-Israel war on Iran will end soon, after oil prices jumped and the growing regional conflict continued to shake markets. After a wave of heavy bombardments throughout Iran, Secretary of War Pete Hegseth promised another round, “The most fighters, the most bombers, the most strikes.” 

“Hegseth has, yes, said that it’s going to be basically death and destruction from the air, and they’re delivering that,” Hooman Majd, an Iranian American writer and journalist, tells The Intercept Briefing. 

“Killing civilians is a hallmark of American air war. This particular campaign Operation Epic Fury is set apart by the relentlessness of the attacks,” adds Nick Turse, senior reporter for The Intercept. “The two militaries — U.S. and Israel — combined were striking a conservative estimate of 1,000 targets per day in the first days of the conflict. Around 4,000 targets were hit in the first 100 hours of the campaign. For another point of comparison, Israeli attacks in the recent Gaza war were also relentless, but this far outpaces the Israeli campaign by more than double the number of strikes.” On Wednesday, Trump told Axios the war would end soon because there’s “practically nothing left to target.”

This week on the The Intercept Briefing, host Akela Lacy talked to Majd and Turse about the latest developments in the U.S. and Israel war on Iran and the growing number of conflicts the U.S. is engaged in. Senior technology reporter Sam Biddle also joined to discuss how artificial intelligence is being used in various U.S. conflicts.

OpenAI on Surveillance and Autonomous Killings: You’re Going to Have to Trust Us

“Airstrikes, air war generally is already so prone to killing innocent people even when you take your time. But whenever you try to hurry for the sake of hurrying — and AI is great at enabling that — you just increase over and over again the chance of killing someone that you didn’t intend to or didn’t care enough to avoid killing,” says Biddle. “So I think that is an immense risk of just accelerating the metabolism of killing from the air by drone, by airplane — with the stamp of ‘intelligence’ that these AI companies are really pushing.”

Listen to the full conversation of The Intercept Briefing on Apple Podcasts, Spotify, YouTube, or wherever you listen.  

Akela Lacy: Welcome to The Intercept Briefing, I’m Akela Lacy, senior politics reporter at The Intercept.

Sam Biddle: And I’m Sam Biddle, senior technology reporter at The Intercept.

AL: Sam, this is your first time on The Intercept Briefing, correct? 

SB: It is. I’ve been at the Intercept for 10 years. I finally got the call. I’m excited.

Akela Lacy: Welcome, we’re very glad to have you. 

SB: Thank you so much.

AL: On a serious note, as we speak, the U.S. is engaged in war and acts of aggression on multiple fronts from the Middle East to the Caribbean and Central America. You have been doing some really important reporting on how the Pentagon is using artificial intelligence in wars and surveillance around the world.

Last month, the Wall Street Journal reported that Claude, an AI tool from the company Anthropic, was used to capture now former Venezuelan President Nicolás Maduro, which set off a dispute between the company and the U.S. government, and opened the door for Anthropic’s rival to swoop in. The Wall Street Journal also reported that Trump has used those same tools in strikes on Iran. Tell us more. 

SB: So what’s been reported is that the Pentagon has made use of a system it has called the Maven Smart System, which is operated by Palantir, the semi-infamous data mining firm. We know based on multiple reports at this point that they’re using the Maven system to essentially accelerate the selection and subsequent destruction of targets on the ground.

This is a way of executing airstrikes at a greater speed potentially, not necessarily more intelligently or with greater accuracy, but I think just faster. And I think people at the Pentagon would probably say, more effectively, more efficiently finding things to destroy and people to kill.

“Target selection is a labor-intensive task.”

“Target selection is a labor-intensive task.”

Target selection is a labor-intensive task. If you can have an LLM like Anthropic’s Claude system — we’ve all seen how quickly they can generate a huge wall of text, of questionable accuracy — can bring that same hyper-speed to creating lists of buildings to destroy and people to kill. I think that is proven to be the biggest value — not just to our military, but to militaries abroad as well.

AL: Sam, what do we know about how the Pentagon is using AI tools in the Trump administration’s various wars?

SB: Under Secretary of Defense Pete Hegseth, there has been a huge, very aggressive push to integrate AI really wherever and whenever possible.

I think that you’re seeing the Pentagon under Hegseth mimic a lot of tech industry rhetoric, which is “we don’t totally understand this technology. We don’t totally know where it’s got to be useful, but we need to use it as much as possible anyway.” I think that you’ve seen DOD under Hegseth be extremely aggressive in the cadence of airstrikes.

This is a Pentagon that believes in killing people. I think, at times, it seems to sort of give itself things to tweet about. This is a political movement and an ideology guiding the Pentagon that I think relishes violence. These AI systems, when you want to blow things up and kill people, these tools can provide a very rapid, turnkey means of having a list of people and places to destroy.

So what we know based on a recent Washington Post report that was discussing the use of Anthropic’s Claude system in Iran, was that it was not just used for target selection, but also target prioritization: Here are the most important targets to attack. Also, something that the Post described as sort of simulating battlefield outcomes. It’s a little unclear what exactly that means. One can imagine just asking a chatbot to basically create a story about how an airstrike could play out. That’s essentially what an LLM does, is generate text that’s plausible based on the inputs. How exactly these simulations are playing out of what value they are, how accurate they are in terms of what might actually happen subsequently in real life is unknown.

“This is a Pentagon that believes in killing people.”

“This is a Pentagon that believes in killing people.”

To me and for the public, the most concerning aspect of what’s been reported about the ongoing use of these LLMs by the Pentagon is the focus on speed. Airstrikes, air war generally is already so prone to killing innocent people even when you take your time. But whenever you try to hurry for the sake of hurrying, and AI is great at enabling that, you just increase over and over and over again the chance of killing someone that you didn’t intend to or didn’t care enough to avoid killing.

So I think that is an immense risk of just accelerating the metabolism of killing from the air by drone, by airplane — with the stamp of “intelligence” that these AI companies are really pushing. If you blow up a school because Claude told you that it was actually an IED factory or whatever, you could say, “Oh, well, the super-smart computer told me to.”

AL: It was the robot. It wasn’t me.

SB: Exactly. We’ve spent the past several years having the tech industry tell us how ultra-smart, ultra-intelligent these systems are. That’s worrying enough when we’re asking them to write our emails for us and do our homework for us. But again, this is the business of killing people. Mistakes are not just mistakes. I think that is now just the way wars are going to be fought, and that is a very troubling new reality.

“This is the business of killing people. Mistakes are not just mistakes. I think that is now just the way wars are going to be fought, and that is a very troubling new reality.”

“This is the business of killing people. Mistakes are not just mistakes. I think that is now just the way wars are going to be fought, and that is a very troubling new reality.”

AL: Backing up a little bit. There is a fight right now between these companies and the government over how, if at all, their tools should be used. We know that they are being used.

But can you tell us a little bit about what is in dispute here? It also sounds like there’s some talk about guardrails being put in place, but we know that means very little in this context. Can you walk us through that?

SB: So the original controversy here was Anthropic, a leading rival of OpenAI. Some would say they have a better product at this point. They got into a dispute with the Pentagon over selling access to Claude, which is their AI chatbot system, akin to ChatGPT.

AL: But it has a human name.

SB: It does have a human name. Don’t you love that?

The company says that they did not want to permit the Department of Defense to use Claude for domestic surveillance of Americans and for killing people without human oversight. The Pentagon says this is woke nonsense, you’re now banned from doing work with the government —and then OpenAI enters.

AL: I will also note in 2024, The Intercept sued OpenAI in federal court over the company’s use of copyrighted articles to train its chatbot ChatGPT. The case is ongoing.

SB: And this is where it gets very strange because OpenAI claims to have the same red lines as Anthropic, but somehow was able to seal a deal with the Pentagon.

Both are very muddled when it comes to what they actually refuse to do. They seem to both want to say that, look, we’re not going to do anything illegal and we’re also not going to engage in these acts — autonomous killing and domestic surveillance — which are largely considered legal.

“It ultimately comes down to what they, what their lawyers decide is legal.”

“It ultimately comes down to what they, what their lawyers decide is legal.”

Appealing to the law is no protection against these acts that the companies are saying that they will not facilitate. I wrote in a piece a few days ago, I think, ultimately, without being able to review the actual contract language for ourselves and to have lawyers go through it carefully, it all just comes down to whether or not you trust the corporate leadership of OpenAI and Anthropic, as well as Pete Hegseth and the White House. It ultimately comes down to what they, what their lawyers decide is legal. We’ve seen White House lawyers say a lot of things are legal: NSA spying, torture, et cetera. So that appeal to the law by these companies is not as reassuring as they want the public to believe it is.

Just one note though: Even though Anthropic’s deal with the Pentagon fell apart, the DOD is still able to use their technology through — it gets a little complicated here — Palantir’s Maven Smart System software, which has Claude in it as a feature, rather than getting it straight from Anthropic.

When you see headlines about Anthropic being banned or being rejected by the military, DOD can still use their software. It’s a pretty nice loophole. So they are still very much in use.

AL: I’ll also mention that the U.S.–Israel war on Iran is also the first example of countries attacking data centers as an act of war, which Sam, you have some reporting coming out on in the future, so everyone look out for that. 

So to recap, the Trump administration appears to be at war with the world. The self-proclaimed “president of peace” has sent U.S. forces jumping from conflict to conflict from Venezuela to Iran to Ecuador and more. As our colleague Nick Turse, senior reporter for The Intercept, tells me on the podcast today, the U.S. has launched attacks in eight countries and killed civilians in two bodies of water — and made threats against five other nations. We also speak with Hooman Majd, an Iranian American journalist and contributor to NBC News, about the latest developments in the U.S. and Israel’s war on Iran, which is ricocheting around the globe. This is our conversation. 

Nick and Hooman, welcome to The Intercept Briefing 

Hooman Majd: Thank you. 

Nick Turse: Thanks for having me on.

AL: Hooman, the Israel–U.S. war on Iran is stretching into another week. A new round of air bombardments hit throughout the country, Al Jazeera reported Monday evening, “We can say this is by far one of the most heavily intense nights in Tehran in terms of air bombardment.” Secretary of War Pete Hegseth promised, “The most fighters, the most bombers, the most strikes.” Israeli Prime Minister Benjamin Netanyahu said he hoped the Iranian people would oust the regime. The civilian death toll in Iran has reached about 1,300 people. To start, what are the latest developments, particularly over the last few days? 

HM: Last few days, I mean, it’s heavy bombardment. That’s what it is.

Hegseth has, yes, said........

© The Intercept