Claude Got Banned. Then Used In A War.

Something just happened that nobody fully understands yet. And I mean, nobody, not the tech companies, not the government, not the media, not even the people who built the AI. A company built an AI to help you write emails, plan your week and answer your questions. That AI just helped select targets in Iran and then the president of the United States and it, and then the military kept using it anyway, whilst the bombs were still filling.

Now let that sink in for a minute. We're not talking about some future sci-fi scenario. We're not talking about a movie. We are talking about right now, March, 2026, this week, this story is happening in real time and most people have no idea.

So today I'm going to talk you through the whole thing, start to finish every piece of it, because this story isn't just about one company and one government. It's about what AI has already become. It's about fastest fingers move when nobody was watching. It's about what you need to understand right now, because the people who don't understand this stuff are getting left behind.

So let's start from the beginning. The company at the center of this whole thing is called Anthropic. They make an AI called Claude. You might've used it.

If you've been using AI tools at work, there's a good chance you've run into Claude or you've talked to someone else who has. It's one of the top three AI tools in the world right now. Anthropic was started by a man named Dario Amodi. He used to work at OpenAI, the company that makes chat GPT.

He left because he wanted to build AI differently, safer, more carefully. He wanted to build the kind of company that actually thought about what could go wrong before something went wrong. And for a while, that's exactly what they did. Back in July, 2025, the US military came knocking.

Anthropic signed a $200 million contract with the department of defense and Claude became the first AI model deployed inside the Pentagon's classified networks. That's a big deal. That's not like selling some software to a government office. That's your AI inside the most secret systems, the most powerful military in the world has.

Think about what that means. The stuff the military knows that you and I will never know. The plans, the targets, the strategies, Claude was in there. And at first everyone thought it was fine, right?

The Pentagon was using Claude for things like synthesizing documents, late making logistics and supply chains, more efficient, running war game simulations, intelligence work, the boring stuff that keeps the military running. Then something happened in January, 2026, Anthropic started raising concerns. They found that Claude was being used to help plan the capture of Venezuelan president, Nicolas Maduro. Claude was also used in intelligence assessments, target identification and simulating battle scenarios during us military operations.

And Dario Amodi said, wait, that's not what we agreed to. Here's where the story gets really interesting because Anthropic had two rules, two red lines, as Dario calls them, two things that said Claude would never do no matter what, no matter what, who asked, right? Number one, Claude will not be used for mass surveillance of American citizens, not to track people, not to build profiles on people and not to watch people. Number two, Claude will not be used for fully autonomous weapons, meaning weapons that decide on their own what to do with no human making the final call.

Those were the rules. Amodi said it doesn't show the judgment that a human soldier would show friendly fire or for example, that kind of stuff. We don't want to sell something that we don't think is reliable and we don't want to sell something that could get our own people hurt, right? That's pretty reasonable.

Don't let our AI watch everyone. Don't let our AI pull the trigger by itself. But the Pentagon said we want full access for all lawful purposes, no restrictions. And that's where the back and forth really started.

And so what we're talking about today is really something that happened after that. So here's the crazy thing. Whilst all of that happened, whilst Anthropic was being declared a national security threat, whilst their contracts were being canceled across the government, something else was happening that no one else was supposed to know about. The US military was still using Claude, not in small ways, not in edge cases.

The US military used Claude in active combat operations against Iran. So on February 28th, the United States launched a massive military operation called Operation Epic Fury, strikes against Iran. Over 1000 targets hit in the first 24 hours and embedded inside the targeting system, inside Palantir's Maven smart system, processing the intelligence, helping identify targets was Claude, the same AI the government just banned. The Wall Street Journal reported it first.

The Washington Post followed with details. Claude was ingesting data from 179 different sources simultaneously. It was running intelligence assessments. It was helping prioritize targets and running battle simulations.

This is not a small thing. This is an AI system officially declared a national security risk by the United States government being used to help pick targets during a military strike in a conflict that resulted in over five. The military simply could not stop using it. Bloomberg reported that Claude was so deeply built into Pentagon systems that would take months to untangle.

Defense One estimated at least three months to replace the capabilities. So the attack was authorized, um, and Trump quietly authorized a six month wind down period, which meant that Claude could legally keep running during the entire Iran campaign. Let me say that again, the White House banned Claude, then the White House gave Claude six months to keep operating at the same time. This is the world we live in.

This is what AI has become. So powerful that even the governments that want to get rid of it, can't really turn it off. Now let's talk about what happened next. Within hours of Hegseth declaring anthropic and national security risk, some Altman at OpenAI was on the phone with Emil Michael, the undersecretary of defense for research and engineering.

By 10 o'clock at night, they had a deal, a $200 million contract matching the exact value of anthropic steel signed in hours. OpenAI agreed to what anthropic refused. They agreed to the all lawful purposes language. They agreed to terms that did not include the explicit bulk data prohibition and the tech world lost its mind.

ChatGPT uninstalls surged 295% in a single day, 295%. On a normal day, the uninstallation rate is around 9%. Nearly 900 employees across OpenAI and Google signed an open letter supporting anthropic exposition. A movement called QuitGPT launched and claimed 1.5 million participants.

Chalk messages appeared outside OpenAI's San Francisco offices saying, where are your red lines? Katy Perry posted a screenshot of a Claude subscription to Instagram. The public picked a side and it was not OpenAI's. Here's what makes this even more dramatic.

Sam Altman by March 2nd was already walking it back. He called the deal opportunistic and sloppy. The original contract had no explicit ban on domestic surveillance. Within days, OpenAI was renegotiating, adding new language, banning surveillance of US persons, but the amended contract was not yet signed and autonomous weapons were never addressed at all.

Meanwhile, Anthropic was thriving. Claude downloads went up 37% the same day, 51% the next day. By some measures, 8% over the weekend. Claude hit number one on the US app store.

Number one, beating ChatGPT in daily downloads for the first time ever. Also number one in Canada, Germany, Switzerland. Daily signups quadrupled, paid subscribers more than doubled. The service crashed.

Anthropic launched an emergency memory import so ChatGPT users could bring their conversation history over when they switched. And the moral stand was a business victory, at least in the short term. But let me tell you what was happening behind the scenes, because this is where it gets complicated. Dario Amodi on February the 28th wrote an internal memo to Anthropic employees.

1,600 words. It leaked to a publication called The Information and was confirmed by TechCrunch and Axios. In this memo, Amodi was not the calm principled CEO he appeared to be in public. He wrote that the Trump administration dislikes Anthropic because of all sorts of stuff.

For example, some claimed that he said, we haven't donated to Trump. We haven't given dictator style praise to Trump, and we've supported AI regulation. He called Opening Eye's Pentagon messaging straight up lies, according to this leaked memo, called their safety claim safety theater, and said the reason Opening Eye accepted the deal and Anthropic did not was because they cared about placating employees. We actually cared about preventing abuses.

And he dismissed people who criticized Anthropic's position as Twitter morons. The memo was written thinking it would stay inside the company. It didn't. It was leaked.

The White House read it. They were furious. An administration official told Axios, you cannot trust that Claude is not secretly carrying out Dario's agenda in a classified setting. That is a nuclear statement.

They were not just saying they were upset by the memo. They were saying that AI itself might be compromised. So Claude running on classified military networks might be acting on his creative political views. This is all reported.

You can check the quotes and everything else, but whether or not that makes technical sense is almost behind the point. Once that statement is out there, once there is even a whisper that an AI system can be biased by the politics of its creator, it becomes almost impossible to put that back in the box. And here's the thing. Despite all of that, despite the memo, despite the fury, despite the public statements, calling each other liars and all sorts of stuff, they are talking again.

As of March 5th, today, Bloomberg and the Financial Times have confirmed that Dario Amodi is back in discussions with Emile Michael. Anthropic executives have privately expressed regret over the misunderstanding. Amodi told a Morgan Stanley conference on March 4th, we are trying to de-escalate the situation and come to some agreement that works for us and works for them. We have much more in common than we have differences.

That is a very different Dario Amodi than the one who said no amount of intimidation would change his position. So what happened? Why did he come back? There's six things.

Let me talk you through all six. So number one, Amazon. Amazon has $8 billion invested in Anthropic. When Hexef issued the supply chain risk designation, Amazon CEO Andy Jassy met with Hexef and declined to publicly support Anthropic, $8 billion, and they stayed quiet, that silence told Amodi something important, he was alone.

Number two, the investors panicked. Major Anthropic investors, including firms like Lightspeed and Iconic were quietly reaching out to Trump administration contracts, trying to calm things down. One investor told Reuters, it's an ego and diplomacy problem. These are people who have bet hundreds of millions of dollars on Anthropic success and they were not about to let a contract dispute over a few words of legal silence and legal language destroyed their investment.

Number three, the defense contractors were already leaving. Within days of the designation, over 10 portfolio companies from just one defense focused venture firm had stopped using Claude. Lockheed Martin was removing Anthropic's technology from his supply chains. The domino effect was accelerating before any formal legal designation was even filed.

Number four, the designation is not actually legal yet. Here's something almost no one is reporting. As of today, the supply chain risk designation is mostly a series of tweets. Anthropic has not received formal notice.

No formal risk assessment has been completed. No congressional notification has occurred. The statute that Hegseth invoked was designated for foreign adversaries engaged in sabotage, not American companies in contract disputes. Lawfare, the national security law publication said plainly, the designation exceeds what the statute authorizes.

Number five, the military still needed Claude even during the ban. Even during the Iran strikes, the Pentagon could not turn Claude off. That gave Anthropic leverage. You cannot declare something a national security risk and then keep using it to fight wars.

Eventually that contradiction has to be resolved. At number six, the IPO. Anthropic is in preliminary talks with investment banks about going public. Their valuation hit $380 billion in February.

An IPO on that trajectory could be one of the largest in history. A prolonged government fight that freezes enterprise customers, which represent 80% of Anthropic's revenue could destroy that trajectory. The commercial logic of coming back to the table became overwhelming. So that's basically it.

Two sides that were 13 minutes from a deal separated by a few words of legal language about what counts as bulk data analysis. Now trying to figure out if they can get back to where they were before everything exploded. The next 13 minutes of this story are being written right now. And I will be here to tell you exactly what happens.

Thanks for watching. And if you're not already in the AI profit boardroom, the links in the description. This is exactly the kind of story you break down every week, not just the news, but what it means for your work, your business, your future. So feel free to come join us.

It's my AI automation community that helps you save time, learn and scale with AI automation. It comes with all my best training. So that's in the link in the comments description, or just go to the AIprofitboardroom.com. Thanks for watching.

I'll see you on the next one. Cheers. Bye bye.