Your AI Won't Write Flirty Texts, But It Will Help the Military Kill
Your AI chatbot refuses to help you write a spicy romance scene. It won't generate a suggestive image. Ask it to help with anything remotely adult and you'll get a lecture about safety guidelines.
But target military strikes that kill people? Apparently that's fine.
This week, we learned that Claude AI (the chatbot many people switched to because it seemed "safer" than ChatGPT) was used to help target over 1,000 military strikes in Iran. The same AI that won't let you ask it to roleplay is being used to identify human targets for lethal operations.
Let that sink in for a second.
The Fight Nobody Saw Coming
Anthropic, the company behind Claude, is now in an all-out brawl with the Pentagon. Not because they refused to work with the military at all. They already signed that deal.
The fight is over WHERE the line gets drawn.
Anthropic says they won't allow Claude to be used for domestic mass surveillance of Americans or fully autonomous weapons (where AI decides to kill without a human pushing the button). The Pentagon says that's not good enough. They want full control once they have access to the system.
According to The Guardian, the dispute came to a head when the government demanded the ability to use Claude however they see fit for "national security." Anthropic said no. The Pentagon responded by declaring Anthropic a "supply chain risk" (a label previously reserved for Chinese companies like Huawei).
And OpenAI? They swooped in and cut a deal with the Pentagon within days.
When Your AI's Robotics Chief Quits in Protest
Here's where it gets messy.
Caitlin Kalinowski, who led OpenAI's robotics team, resigned on Saturday. Her reason? The Pentagon deal crossed lines she wasn't comfortable with.
She posted on X: "This wasn't an easy call. AI has an important role in national security. But surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got."
Think about that. A senior leader at one of the world's most powerful AI companies quit because she thought the military deal happened too fast and without enough ethical consideration.
Even OpenAI's CEO Sam Altman admitted the timing looked "opportunistic and sloppy."
The Ethics Are Backwards
Here's what makes this so strange.
AI companies spend enormous resources making sure their chatbots won't:
- Write anything sexual
- Help you cheat on homework
- Generate fake news
- Create deepfake images of celebrities
These are real harms, sure. But the resources poured into preventing a college student from using AI to write an essay seem wildly disproportionate when the same companies are handing over tools that help decide who lives and dies.
As one Reddit user put it: "Anthropic doesn't allow erotic conversations but allows lethal targeting. What does that tell you about their priorities?"
It tells you the "safety" these companies care about isn't YOUR safety. It's liability protection and brand management.
Why This Affects You
You might think military AI has nothing to do with your daily ChatGPT use. But it does.
First, it shows who actually controls the AI you use every day. You don't. Not even close. These tools can be repurposed for uses you'd never agree to, and you have zero say in it.
Second, it reveals what AI companies actually optimize for. They'll block you from doing harmless creative writing to avoid controversy, but they'll sign deals to help governments kill people if the check is big enough.
Third, once AI gets deployed in classified military systems, the companies lose all visibility into how it's being used. As Sarah Kreps, a Cornell professor who previously served in the Air Force, told The Guardian: "Once you hand this over to the military, you no longer need Anthropic's approval to use it as you see fit. It goes into not just a black box, but Black Ops."
The Pattern You Need to Notice
This isn't the first time we've seen this pattern.
Tech companies build consumer products. They market them as safe, ethical, responsible. They put guardrails on what YOU can do with them.
Then they quietly sell access to governments and corporations with far fewer restrictions.
Remember when Apple refused to unlock the San Bernardino shooter's iPhone for the FBI, citing user privacy? That was 2016. A principled stand that earned them massive public goodwill.
Fast forward to 2026 and we have AI companies that won't let you write a love letter but will help the military identify bombing targets.
The difference? Apple was protecting user privacy. These AI companies are restricting user AGENCY while expanding government power.
What Happens Next
Anthropic says they'll challenge the Pentagon's "supply chain risk" designation in court. That's a fight worth watching because it could set the precedent for whether AI companies can draw ethical lines or if national security demands will always win.
Meanwhile, OpenAI now has the Pentagon contract Anthropic walked away from. Whether that makes them more pragmatic or just more willing to compromise is up for interpretation.
And for the rest of us? We're left using AI tools where we can't write a slightly suggestive poem, but somewhere in a classified system, that same technology is helping make life-and-death decisions.
The message is clear: AI companies will protect you from mild embarrassment, but not from surveillance. They'll stop you from creating art they don't like, but they'll help governments kill more efficiently.
You just have to hope you're never on the wrong side of that targeting algorithm.
What You Can Do
You can't control what the Pentagon does with AI. But you can pay attention to which companies are willing to say no.
Anthropic drew a line, even if it was later than they should have. That matters.
Watch how AI companies respond when faced with these choices. The ones that prioritize your safety over government contracts are the ones worth supporting.
And maybe think twice before trusting a company that censors your creative writing while helping the military decide who to kill.