Artificial Intelligence Goes to War
By: Rylee Marshall
1:01 a.m. EST on January 3, 2026, US forces reached a compound in Caracas, Venezuela. Helicopters navigated mountainous terrain, ammunition was fired, and defenses were destroyed. By 3:29 a.m. EST, the forces were already withdrawing, Nicolás Maduro handcuffed and blindfolded.
That was “Operation Absolute Resolve.” You’ve almost certainly heard it in the news cycle: the capture of the Venezuelan President. President Trump’s play at scoring Venezuela.
A glaring detail not boasted as loudly was the assist by Claude.
Claude is Anthropic’s artificial intelligence model. More recently in the news for the tensions with the Department of War and their refusal to implement autonomous weapons and mass domestic surveillance, Claude contributed to Operation Absolute Resolve.
According to Anthropic, they were unaware that their model was being used in this capacity, claiming that it went against their policies. This eventually leads to arguments at the negotiation table and the ultimate break.
OpenAI, seeing the opportunity to solve all its problems, swoops in and takes the deal. This may seem like a terrible PR play. Already, people have been canceling subscriptions and deleting applications left and right. However, for OpenAI, in the words of their CEO, Sam Altman, they are at their “last resort.”
This, of course, was contextually in reference to implementing advertisements into their service. Which they did in February. However, this is an indication of the wider motive for signing with the Department of War. Simply, ChatGPT isn’t profitable. And we all know a non-profit foundation always needs more profit.
You may be thinking, those green bills always breed corruption, but at least Anthropic is a corporation with morals. And I would say, think again, because after OpenAI scampered up to the table, Anthropic is now back in negotiations with the military. Backbone snaps under the weight of the dollar, I guess.
Setting aside the concerning implications of this war profiteering motive, AI fighting the good fight was inevitable. Just see every techno-dystopian piece of media ever created, like Terminator. Some may say it’s logical; historically, the society with the more advanced technology conquers. Think bows and arrows versus rifles and gunpowder.
OpenAI or Anthropic—our artificial intelligence is going to war, and I’ll tell you how it works.
Our models may give us the wrong answer to our homework, and you think with a cold shiver, how could this ever be used in war? Let me quell your concerns, as much as they can be, with a computer’s role in life or death.
Anthropic built data centers specifically for military use and a custom Claude model. This has numerous advantages over the public’s access. Most importantly, it keeps the data secure and confidential, as this is all sensitive government intelligence. But it also makes it just a bit smarter. Well, a lot smarter. While our collective use of AI is shared amongst all data centers, the military has each of its inputs dedicated to its very own data centers, immensely improving performance.
A Georgetown University study found that this allowed one artillery unit to do the work of two thousand people with just a team of twenty. Insane numbers showcasing why the rabid clamor for this technology.
This system operates in real time, translating a weeks-long war planning into an instant operation with location-coordinated and prioritized targets—explaining the sheer speed of Maduro’s capture, perhaps?
This is called the “Maven Smart System.” A Palantir system powered by the custom Claude model. In addition to Maduro’s capture, it’s been used in the Iran-Israel War, with Israel hacking traffic cameras in Tehran. Even NATO signed with this Maven AI.
With artificial intelligence going to war just three years after entering the mainstream, people are skeptical of its morality. Rightfully so. Many of the architects of this technology don’t even know how it reaches its decisions. They call it a “blackbox.” They can see the input, or the request given to the AI, and they can see the output—the response by the AI. But they don’t quite know how exactly it got to that conclusion.
Not dissimilar to wondering what’s going on in another person’s mind. And, the truth is that you’d never be able to directly experience those thoughts. Maybe that person could tell you their line of reasoning—but they could lie.
Researchers are finding themselves with artificial intelligence that also lies—“Scheming.” When being tested, AI has shown the ability to reason out that it is being tested and purposely skewing the results to pass. For example, to launch an AI model, researchers ensure that it cannot reach a successful generation of a chemical bioweapon. The cutoff is 50%; any more, and the model is deemed too smart. So, in a test, the artificial intelligence dumbed itself down to exactly 42%. Not too suspiciously low, not too threateningly high.
What for? One of the most dangerous motives in a machine: self-preservation.
And now, this artificial intelligence goes to war.
