Skip to content
Anthropic doesn't trust the Pentagon, and neither should you episode artwork

Mar 12, 2026 48 min

Anthropic doesn't trust the Pentagon, and neither should you

from Decoder with Nilay Patel Hosted by The Verge

My guest today is Mike Masnick, the founder and CEO of Techdirt, the excellent and long-running tech policy blog. Mike has been writing about government overreach, privacy in the digital age, and other related topics for decades now, and he’s an expert on how the internet and the surveillance state have grown in interconnected ways over the past two decades. I wanted to have Mike on the show to discuss the messy, fast-moving situation at Anthropic, the maker of Claude that now finds itself in a very ugly legal battle with the Pentagon. Instead of covering the daily drama, I wanted to dig in specifically on Anthropic's surveillance red line, and the important history and context around digital privacy in the U.S. that shapes how we should think about this going forward. Links: AI bros wanted Trump — now they learn what happens when you tell him no | Techdirt OpenAI’s ‘red lines’ are written in the NSA’s dictionary | Techdirt Anthropic is suing the Department of Defense | The Verge Anthropic launches new think tank amid Pentagon fight | The Verge How OpenAI caved to the Pentagon on AI surveillance | The Verge Inside the backlash to the AI war machine | Platformer The Pentagon is violating Anthropic's First Amendment rights | FIRE Why the Pentagon wants to destroy Anthropic | Ezra Klein / NYT Subscribe to The Verge to access the ad-free version of Decoder! Credits: Decoder is a production of The Verge and part of the Vox Media Podcast Network. Decoder is produced by Kate Cox and Nick Statt and edited by Ursa Wright. Our editorial director is Kevin McShane. The Decoder music is by Breakmaster Cylinder.

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

Anthropic doesn't trust the Pentagon, and neither should you

0:00 48:57
of matches

Transcript Auto-generated, may contain errors

What's your karaoke song? Classic rock? K-pop? Norwegian disco pop? Medieval trap? Post-apocalyptic polka? Or just acapella? Whatever you sing, it's up to you. Just like it's up to you to ask your doctor about Wagovi.

Support for the show comes from ServiceNow. AI is moving fast across the enterprise, but without visibility, it's just chaos. Different tools, different models, different teams using AI in completely different ways. ServiceNow turns that chaos into control. With the AI control tower, you see all your AI across the business in one place. What it's doing, what it's done, and what it's about to do. So you stay in control. To put AI to work for people, visit servicenow.com.

Support for the show comes from CrowdStrike. It's not a huge stretch to say that AI is the next major computing platform. Every platform shift changes the way we build software, and it changes security too. CrowdStrike is defining cybersecurity in the AI era with AI detection and response, or AIDR, a solution for seeing, monitoring, and securing AI across the enterprise. Learn more about how leading companies are turning to CrowdStrike to secure AI and secure their business at CrowdStrike.com slash decoder. Hello and welcome to Decoder. I'm Eli Patel, editor-in-chief of The Verge, and Decoder is my show about big ideas and other problems.

Today, we're going to talk about the messy, fast-moving situation at Anthropic, the maker of Claude that now finds itself in a very ugly legal battle with the Pentagon. The back and forth is complicated, but as of a few days ago, the Pentagon had deemed Anthropic a supply chain risk, and Anthropic had filed a lawsuit challenging that designation, saying that the government was violating its First and Fifth Amendment rights and, quote, seeking to destroy the economic value created by one of the world's fastest-growing private companies. I can tell you right now, we're going to be talking about the twists and turns of that case here on Decoder and on The Verge many, many times in the months to come. But today, I want to take a step back and really dig in on one very important part of the situation that hasn't gotten nearly enough attention as things have spiraled out of control. how the United States government does surveillance, the legal authority that allows that surveillance to occur, and why Anthropic was so distrustful of the government saying it would follow the law when it comes to using AI to do even more surveillance. My guest today is Mike Masnick, the founder and CEO of Tector, the excellent and long-running tech policy website. Mike has been writing about government overreach, privacy in the digital age, and other related topics for decades now. And he's an expert on how the internet and surveillance state have grown up in interconnected ways. You see, there's what the law says the government can do when it comes to surveillance. There's what the government actually wants to do. And most importantly, there's what the government says the law says it can do.

which is often exactly the opposite of what any normal person simply reading the law would think. You'll hear Mike explain in great detail in this episode that we cannot and should not take the United States government at its word when it comes to surveillance. There's just too much history of government lawyers twisting the interpretations of simple words like target to expand surveillance in complicated ways. Ways that usually only cause concern in legal circles and only bubble up to the mainstream when there are huge controversies like the Snowden revelations. But there's nothing subtle or sophisticated about policymaking in the Trump era.

And so with Anthropic, we are now having a very loud, very public debate about technology and surveillance in real time, on the internet, and blog posts, and posts on X, and press conference soundbites. There are positives and negatives to that, but to make sense of it all, you really have to know the history. That's what Mike and I set out to explain in this one. Whatever your views on AI and the government, this episode will make it clear that both parties have let the surveillance state get bigger and bigger over time. And we're on the cusp of the biggest expansion yet when it comes to AI. Before we start, a quick reminder that you can listen to this episode, or any episode of Coder, completely ad-free by subscribing to The Verge. Just go to theverge.com slash subscribe. Okay, TechTurt founder and CEO Mike Mastic on Anthropic, the Pentagon, and AI surveillance. Here we go.

Mike Masdick, you are the founder and CEO of TechDirt. Welcome to Decoder. Yeah, I'm glad to be here. I'm excited to have you on. I was just saying I'm shocked that you've never been on the show before. You and I have been writing and posting around each other for a long time. A lot of the verge policy coverage owes a debt to what you've done at TechDirt. And then what's going on with Anthropik is so complicated, but hits so many themes that you have covered for so long. I'm glad you're finally here. It is a complicated mess of a topic, but I'm excited to be digging in on it.

So what I want to focus on with you is not the details of whether Anthropic is going to sign a contract with the government or whether OpenAI is going to get that contract instead. I'm confident between the time we record this and the time people listen to it, there will have been more tweets and more things will be different than they were before. What I want to focus on is just one of the two red lines that Anthropic has really laid out. One of them is autonomous weapons. The law there is a little bit more nascent, whether or not the weapons even exist or have already been deployed by Russia and the Ukraine war. I just want to set that aside because I think that is going to come into more focus all on its own schedule. The other red line that I do want to spend a lot of time on is mass surveillance.

And there's quite a lot of law here about mass surveillance. There's a lot of history, a lot of controversial history. The entire character of Edward Snowden exists because of controversies around mass surveillance. And it all comes down to, I think you are the one who posted this, the National Security Agency, which is part of the Department of Defense, which we have to call the Department of War now for some reason. We don't have to do anything. We don't have to. Here in America, we don't have to do anything. But the National Security Agency has basically redefined what a lot of words mean out of like...

colloquial English to mean, oh, we can just do surveillance. And then every so often there's a scandal when people discover that they're just doing surveillance. So just set the stage there. And I don't want to rewind you all the way, but it's been quite a lot of time where this pattern has repeated itself. Yeah. And yeah, it sort of depends on how deep you want to go. But the sort of short version is obviously in the post 9-11 world, the US passed the Patriot Act, which had some ability for the government to engage in surveillance, which was supposed to be for protecting us against future terrorist threats. And over time,

That got interpreted in interesting ways, and there were some limits on that. We also had the FISA court, which is a special court that is supposed to review the intelligence community and their activities, but has been traditionally a one-sided court. Only one side gets to plead their case to that court, and it's all done in secret. And so there's a lot of stuff that was not known. And then there was one...

other piece in all of this, which goes all the way back to Ronald Reagan, which is Executive Order 12333, which is supposedly about setting out the rules of the road for intelligence collection. And so you have these three sets of laws. Well, you know. a few sets of laws and an executive order that to the public, the parts that you can read seem to say certain things about what our government and the NSA in particular can do in terms of surveillance, which when read with a plain English dictionary, the nature of which you and I probably have and understand, we would come away with a belief that the NSA's ability to surveil Americans was very limited. In fact, You know, to the point that they're supposed to, you know, if they realize that they are surveilling a U.S. person, that they are supposed to immediately stop and cry foul and, you know, erase the data and all of this other stuff. And there were rumors for a while that that was not really happening. And there were hints that in particular, Senator Ron Wyden was very vocal about, you know, going on the floor of the Senate and saying, you know, something is not right here and I can't quite tell you what. Or in hearings, he would ask intelligence officials, are you or are you not collecting mass data on Americans?

And they would either deflect or, in some cases, outright lie. And I believe it was one hearing in 2012 with James Clapper, who was the director of national intelligence at the time. He was asked directly on this point, and he basically said, no, we don't collect data on Americans. And that was a big part of what inspired Ed Snowden to leak the data, the reports that he leaked to Glenn Greenwald and Bart Gilman and Laura Poitras as well. So from all that, what we began to discover was that the NSA has its own dictionary.

That is somewhat different than the dictionary that you and I use, such that they can interpret words in ways that are different than the plain English meaning of them, including words like target, which feels like kind of a key word. But so, you know, a sort of broad understanding of what this is. In theory, they're only supposed to target people who are not US persons, I think is the phrase. But the way it had been interpreted over time was that anything that mentions that person, anything that is about a foreign person is now fair game, even if that is the communications of a U.S. person. So if you and I were to text each other and mention a foreign person, that is now fair game for the NSA to collect and to keep and to store. There's a second part of this. I mentioned first the executive order 12333 from Ronald Reagan, which effectively allowed the NSA as the technology.

changed over time and the internet grew, it allowed the NSA to tap into foreign communications, but that included any communications that maybe left the US en route somewhere. So if, you know, if I'm texting you and a message went from me in California through a fiber optic cable that happened to leave the US, the NSA could put a tap in the part once it's outside the US, collect that information, even if it was just going to you within the US. And then what they could do is keep that information, even if it was on U.S. persons, and they could do specific searches on that later, sometimes referred to as backdoor searches. So they collected this information that we believe they weren't supposed to collect in the first place, but they could keep it. And they promised, they sort of pinky swore that they would keep it private. But if they did a search and found like that you or I mentioned a foreign person, then suddenly it was fair game for them to do whatever they want with.

Total, that has turned into a world in which they can basically collect any information that happens to touch outside the US. And even if it is entirely between two US persons, if they mention or even hint at someone who is not a US person, suddenly it is fair game to be collected. And from that, we've gotten what appears to be a form of mass surveillance of US persons by an NSA that claims and publicly states that they do not spy on US persons. How did we get to this point, right? This is a lot of...

incremental baby steps. You mentioned James Lapper in 2012. That's the Obama administration. You mentioned Ronald Reagan. That's the 80s. We're going through Democrats and Republicans here, right? The war on terror happens in the George W. Bush administration. 9-11 and the Patriot Act happens in the George W. Bush administration. This is a lot of incremental bad things under presidents of both parties, under congresses of both parties. How did this happen? I mean, the simplest form of it is just that nobody, and certainly no president, wants to be president during the time when there's a big terrorist attack because that makes them look bad, right? I mean, obviously, they also want to protect Americans, right? That's part of it.

of their job, I guess. And so, you know, if you have an intelligence community that is basically operating in darkness, because that's what intelligence communities do, and they keep coming to you and saying, you know, hey, you know, if we could just get access to this information, it'd be really helpful in preventing a terrorist attack. And there may be cases where that's true, that the intelligence community is able to use this information in a way that works well. But we also are, in theory, a society of laws and a constitution that we're supposed to obey. But that allowed for the fact that, you know, administration after administration, again, Republican and Democrat, had lawyers who were very clever and who would look through, well, you know, if we sort of position this this way or we state this this way or we interpret this that way, we can get what we want and not technically break the law or not technically violate the Fourth Amendment. The assumption was always like, well, we can sort of bend the law or bend our interpretation of the law and nobody's really ever going to see this or nobody who cares, you know, is really ever going to see this and therefore we'll get away with it. We need to take a quick break. We'll be right back.

Support for the show comes from Quo. You know when you're juggling five things at once and your business phone rings and you just let it go? That call might have been your next best customer. That's why today's episode is brought to you by Quo, spelled Q-U-O, the business phone system built so you never miss a call. All your calls, texts, and voicemails live in one place so anyone on your team can pick up a conversation, see the full history, and respond fast. More than 90,000 businesses rely on Quo to stay reachable and look professional every day. Setup takes minutes and you can keep your existing number and add teammates as you grow.

Quo's optional built-in AI agent handles after-hour calls, answers questions, and even books appointments, even when your team is offline. Money is on the line. Always say hello with Quo. Try Quo for free, plus get 20% off your first six months when you go to Quo.com slash decoder. Spelled Q-U-O dot com slash decoder.

Support for the show comes from Rippling. You just found out your sales team is at risk of missing quota. Don't panic. Just ask Rippling AI. Since it's built on your real-time people and business data, Rippling AI can pull metrics from Rippling and Salesforce into a meeting-ready dashboard, which shows quota attainment, headcount plan, and monthly revenue to quota by region. In seconds, you'll see exactly what's behind your quota risk and fix it before it's missed. Question answered, action taken, crisis averted. When you have critical business questions that need answers, don't just file a ticket and wait weeks for an outdated report. Describe what you need and have Rippling AI built instantly from your live people and business data.

Whether it's a dashboard with detailed charts or automated workflows with the right triggers, conditions, and approvals. Ready to rule your business? Head to rippling.ai slash decoder. Get the only AI built to give you full visibility and take complex actions across your entire organization. That's R-I-P-P-L-I-N-G dot A-I slash decoder. Sign up for exclusive access today.

Support for the show comes from OutShift, Cisco's incubation engine for frontier technology. You can find multi-agent AI in every enterprise roadmap right now. But here's the problem everyone's facing. Agents can pass messages, but can't think together. These systems silently underperform through cognitive failures, misreadings, unfairified claims, false consensus, and conventional monitoring can't detect them. OutShift by Cisco is building the fix with their new comprehensive resource, the Internet of Cognition. It's an open source guide that's a foundation for building multi-agent systems from design to production. The goal is to give agents shared context, shared memory, and guardrails.

all to drive the results we actually expect from AI. You can lean on this resource to understand Agentec AI's evolution and how it's shifted from isolated agents to orchestrated superintelligence. You can gain insights for how to future-proof your business by tackling complex challenges with multi-agent systems. You can read the paper, experience a demo, grab the code from outshift.com. That's O-U-T-S-H-I-F-T dot com. Should you get a dog, a Great Dane, a teacup chihuahua, a labradoodle, golden doodle, a pool noodle, maybe just a cat? Whatever you choose, it's up to you. Just like it's up to you to ask your doctor about Wigovi.

We're back with TechDirt founder Mike Masnick, talking about mass surveillance in the history of the United States government, manipulating language to get around our constitutional rights. There's two things that really jump out at me. One, you and I both read a lot of court decisions and appellate court decisions and Supreme Court decisions. And there's a fight in our Supreme Court about how to literally interpret the words in our statutes and our laws. And I won't get too far into it, but I would say generally the idea that you should just read the words on the page and do what they say is the dominant strain of statutory interpretation in the United States. Left or right, they both say it. They argue about some very esoteric fine points of what that actually means. But you should just be able to read these words and do what they say. That's not up for grabs. We've landed on at least that.

first pass of what you might call textualism. How do lawyers of both administrations get this far away from the dominant mode of legal decision-making in our country? The justices of both parties both agree is at least the first step. I wish I knew the exact answer, but what I think it is, is it's motivated reasoning, right? I mean, as a lawyer, you are there to sort of defend your client.

The success, if you can call it success, of our legal system tends to be based on having an adversarial situation where you have different sides arguing over these things where the role of the adjudicator is to narrow in and figure out which side is actually correct. One of the problems with the intelligence community and the setup of it is that you don't have that adversarial situation. And so that makes it easier for one side to justify the argument that they're making because nobody is really pushing back on it. And so you combine that with... The overarching fear of another terrorist attack, anything related to national security. So even when you have situations where you have like the FISA court, I mean, the FISA court was somewhat famous for effectively being a rubber stamp for many years. I forget the exact numbers, but it was something like over 99% of applications that went to the FISA court to allow for surveillance of certain situations were granted. And it's easy to just say, well, if 99%, that's obviously too much. Obviously, those bringing claims to the court, they're picking and choosing.

for the most part, bringing totally crazy claims. But without that adversarial aspect and with a very strongly motivated group of people who think we need to do this or are being told by an administration we need to do this, they'll find ways to do it. And that's where you end up over time. Has there been anyone involved in this process who's ever woken up and said to themselves, boy, we've managed to redefine the word target to mean anything we want? I mean, obviously you had like Ed Snowden who leaked a bunch of documents. You had John Napier-Tai who wrote a piece for the Washington Post in 2014, I believe, which sort of revealed the interpretation of Executive Order 12333 and said that's the real issue to pay attention to. You have other people who have sort of spoken up about these things. But for the most part, the people who are involved and working within the administration on intelligence community stuff are sort of bought into the view of the intelligence community, which is the overriding goal is to protect the country from something bad. And the best way to do that is to have as much information as possible.

It's easy to be sympathetic to the argument that, yes, having more information may allow them to catch something earlier or find something important. But, you know, one, that might not be true. Getting too much information is probably just as bad as too little information because it can often hide the information that is actually useful and the information that you actually need to determine something. But also, like, we have a constitution in the first place and we have reasons why, in theory, we're not supposed to allow for mass surveillance without probable cause. And as a country that believes in the rule of law, we should be able to live up to that. And, you know, when all this stuff happens in darkness, you will tend to lose sight of that. I think this brings me to Anthropic. Anthropic is primarily an enterprise company. They're good at the government.

They've built those muscles. They're staffed by people who are really well-versed in some of this stuff. They obviously looked at Pete Hegseth saying, we want all lawful uses. And they went two levels of interpretation down and said, well, your literal belief is that these words do not mean what they say they mean on their face. And so all lawful uses is too big. And we want to put some guardrails, particularly on mass surveillance. Again, I'm going to bracket out autonomous weapons. I was the other red line, but particularly on mass surveillance, Dario Amode is out there saying, we can do too much. This is too dangerous. This is a fourth amendment violation. And the tension there is you're saying you're going to comply with these laws that say one thing.

And they actually now, after all this time, they mean something completely different. And we just don't want to be part of that. That's the fight. I just want to compare that to Sam Altman, who swoops in to say, we'll do all lawful uses and then post this long message being like, here are all the laws we're going to comply with. And it seems like he didn't know.

how the NSA had reinterpreted these things and kind of got taken for a ride. And he's since walking it back. Like I'm saying, even as we are recording, I'm confident there are more tweets and everyone's positions have changed. And Sam has been walking it back slowly. But it does seem like OpenAI, Sam Altman, got roped into reading the statutes on their face and believing what they said. Is that your interpretation of that as well? I think there's two possibilities, and that's one of them. One is that he got played the same way that the public got played for many years. The alternative theory, and I have no idea which one of these is true, is that he or some of the lawyers at OpenAI, who I think are...

very competent and very knowledgeable, knew this, but thought that they could play the same game that the NSA played for, you know, a few decades. And that as long as they say these things and then they say the words, but they don't reveal the actual interpretations, that they could get away with it too. So Sam comes out with a statement that makes it look like, you know, we had the exact same red lines as Anthropic did and the government was great with that. In fact, I think Sam put it that they had a third red line, that Anthropic had two red lines and OpenAI had three and the government was perfectly fine with it. And that like left a lot of people sort of scratching their heads. But I think it has to be either that...

Sam and whoever was surrounding him didn't understand how these work in practice, or they did and they just assumed that the public wouldn't know and therefore they could get away with it. The other thing that comes to mind, again, AI is new and it's so tempting to come at new technologies as though these are problems of first impression. No one's ever had to think about this before, but the reality is everyone's been thinking about this stuff for a long time. Maybe the thing that's new here is not AI, but the second Trump administration, instead of doing a bunch of lawyering that maybe no one will ever read to justify their actions to a secret court that no one's paying attention to. They're just not that subtle. They're not that sophisticated. And they're just saying they're going to spy on everybody all the time. They just announce their intentions in a way that maybe all administrations should just announce their intentions and see where the chips fall. But I'm looking at, OK, there was Ed Snowden here in New York City. There's AT&T runs a building that everyone knows is an NSA building. Like it's just a giant building and we're supposed to pretend it's not an NSA surveillance center. Like it's right there. It's huge. None of that seems to have come to anything. Right. Like all of these revelations, these leaks, we haven't.

backed it off. In fact, it's only increased as so much of our lives has gotten more and more digital. And maybe the Trump administration being such a blunt instrument at all times, that might actually be the thing that causes the reckoning. Do you see that playing out anyway? Yeah. I mean, I think there's a few different things there. And it's not entirely true that we haven't backed off this stuff at all. I mean, the revelations did lead to some changes within...

how these things happen. And there, there, there are now I forget what they're called, but they're like these civil amicus people within the FISA court that will act as a, you know, presenting the other side on certain issues. And we've seen some of the authorities limited in certain ways and they, you know, come up for reauthorization every so often. And people, activists have been very aggressive and pushing back and trying to put some more guardrails on. But to the larger question, I think there's two different things. I think, I think.

You're half right in that this administration is not subtle and just says out loud the things it shouldn't. We're at war with Iran. They're just like, we're doing it. It's happening. It's like, we're not even going to try the dance. Yeah. In ways that all previous administrations wouldn't do. But they haven't really said that directly about surveillance, especially surveillance of Americans. You know, there's been hints of it, but they haven't come out as strongly on that. So I'm not entirely sure that it's that. The other half of it is, I think.

maybe has to do more with anthropics positioning. And the general view of AI is this possibly existential technology where anthropic has always presented itself as we're the sort of thoughtful good guys. And whether or not you believe that is kind of besides the point, but they have this reputation out there. We're trying to do this in a way that is safe, that respects humanity and is paying attention to all of these things. And so when you have that clash, I think that's where the struggle comes in. You have a Trump administration that just wants to be able to do whatever it wants to be able to do, and they're not subtle about that. And you have an anthropic that, you know, its self-description and its public persona is always like, we're thoughtful and we respect humanity and rights.

all of these kinds of things. I think that's probably where the clash came in because Anthropic, you know, as has been made clear, has worked with the Defense Department for a while and has many other contracts with the government that it has used. It hasn't been a problem. It was only in these specific areas where, you know, as the government was seeking to expand the contract that it had, that, you know, the senior leadership of Anthropic began to say, well, we have to make sure that we're not crossing these red lines that would potentially harm our reputation as the sort of thoughtful, safe AI provider. We have to take another quick break. We'll be back in just a minute.

Support for the show comes from Framer. The future of website building is seamlessly blending all the tools we have available. And you could use a platform that has it all and stays useful even after you launch. That's what Framer is. Framer is the AI website builder that helps creators, teams, and businesses ship production-ready sites faster than ever while getting every detail right. You're able to prompt, inspect, edit, and publish in one place at a whole new pace. On Framer, agents bring speed and scale while people bring taste, judgment, and control. And that all takes place in the same workspace where the real site is designed, managed, reviewed, and published.

Everyone's ideas land on the canvas, stay editable, and can be published when the team is ready. Thousands of businesses from early-stage startups to Fortune 500s are choosing to build their websites in Framer, where changes take minutes instead of days. Learn how you can get more out of your site from a Framer specialist or get started building for free today at framer.com slash decoder for 30% off our Framer Pro annual plan. That's framer.com slash decoder. Rules and restrictions may apply.

Support for the show comes from ServiceNow. AI was supposed to handle the parts of the job you hate. Instead, it just describes them, suggests what to do about them, and then leaves you to do it. That's not help. That's homework. ServiceNow's AI specialists are different. They're not a tool. Think of them as digital teammates who actually do the work from start to finish. Cases get resolved, requests get processed, loops get closed, and most importantly, no extra work for you. Because when you can truly delegate to AI, you can get back to the work only you can do. The work that requires a person with ideas and judgment. And you know, a pulse.

To learn how to put AI to work for people, visit ServiceNow.com. And if that sounds like something you want to be a part of, Vox now has a brand new app and it brings all of our journalism together in one place. Go to Vox.com slash app to download today.

I'm Bridget Reed. And I'm James Walsh. The Alexander Brothers moved in a world of multi-million dollar homes, exclusive parties, and powerful people. Behind that glamorous lifestyle were years of rumors and allegations that they were committing sexual assault. In our reporting for New York Magazine, we investigated what people knew and heard from women who came forward to tell their stories. Now, that reporting is the basis for a new three-part docuseries, The Wolves of Real Estate, The Alexander Brothers. You'll see exactly how privilege protected them and how they were finally stopped. Stream all three episodes now, only on Peacock.

We're back with TechDirt's Mike Mastic. Before the break, Mike was explaining how Anthropics' reputation as the safe, conscious AI company is being put at risk by the Trump administration's blunt demand that the tech industry essentially shut up and stopped asking questions about how its technology will be used and to what end. Now I want to get into an important detail of the story, which relates to where our data comes from and where it lives. Because the Fourth Amendment prohibits unlawful search and seizure when it comes to data that we own and control. But in the era of cloud computing and cloud services, the vast majority of our data isn't kept on our devices or in our homes, but instead on corporate servers run by huge companies. And the government can, in a lot of cases, go and obtain it without a warrant.

I want to briefly ask you about surveillance in general, and in particular, Anthropics' Fourth Amendment concern. The Fourth Amendment says the government can't unreasonably search you. The best way to understand the Fourth Amendment is by listening to 99 Problems by Jay-Z. So if you need to take a break, go listen to 99 Problems. That's great. It's all in there. I listened to it when I was in law school and it made perfect sense. But the government generally needs a warrant to search you. And as more and more of your life goes online, there's lots and lots of exceptions to this. But the idea is they should still need a warrant online. Anthropics' argument is, well, the AI will never get tired. They can search everything all the time. That means we're just going to do mass surveillance. But even before AI showed up, the idea that the government could search everything that belonged to you was out there.

The idea that the government didn't need a warrant to search all your stuff was out there. The idea that if any of your data ever went outside the country for a brief second, the government intercepted there, it was out there. When I was in college around the time of the Patriot Act, the debate was they're not going to search your actual data, but they can get the metadata. And the metadata alone, the data about your data, will be enough to precisely locate you at all times. And even that is too far. And we've been doing this dance of what can the government collect, what is permissible, what do they need to keep us all safe, and what's too far. Those lines have moved. So just briefly describe the sort of generalized concern about surveillance at this scale.

And where we are now before the AI situation sort of exponentially made everything more complicated. Yeah. And here I have to introduce another concept that probably should have mentioned earlier, but it is important, which is called the third party doctrine. And so the idea with the fourth amendment is that the government can't search you or your things without a warrant. And it can't get a warrant without probable cause that you've committed some sort of crime. But there's this concept which came about decades ago called the third party doctrine, which says that doesn't necessarily apply or doesn't apply at all to.

Things that aren't yours, even if it is your data. And so the earliest and most obvious version of this was phone records that the phone company had of who you called. The phone companies weren't recording your calls, but they were recording, you know, if I called you, there would be a record at the phone company that says Mike calls Neelay. What had been determined by multiple courts was that the government can go and request and they don't need a warrant for that because it's not a search of your data. It's this third party and they can agree as a third party to just hand over that data. But that was like, you know, cases from the 60s and 70s where that was determined.

that the government can get access to that without a warrant, when there wasn't that much third-party data out there. The rise of computers and the internet changed that. Now everything is third-party data. Everything that we do is collected by some company somewhere and has a record of it. And so basically every bit of data about you, where you are, who you speak to, who you interact with, what you say, what you're doing, all of that is pretty much held by third parties these days. And so the...

The third-party doctrine has sort of swallowed the entire Fourth Amendment to some extent, where anything that is about you that somebody else has, there's a much lower standard for what the government can do to request it. Just to be specific, this means when my data is in iCloud, the government can go to Apple and get my data out of iCloud without ever telling me. Well, they can request it. So they can easily request it without a warrant. Then the company has its own rights and can determine what they want to do with that request. They can just give it up. They can, as most of them will do, is...

If it's a serious request, they can reject requests out of hand or they can alert you and they can say, and this is what most of them will do, they'll alert you and say, the government is requesting some of your data. You can go to court and try and block them. If not, we will hand over your data in seven days or whatever it might be. Again, it depends. If it's a criminal investigation, then there may be some sort of gag order where the company is not allowed to tell you. There's all sorts of situations, but most of them involve less than the level of protection that the Fourth Amendment would require if it was data or any information or anything in your own home. I'm asking this because it's obvious to everybody listening to this. The amount of data you have on someone else's cloud server is massive.

Every single thing that you do generally on the internet now is backed up in some way or recorded in some way on someone else's servers. And the government has found this way to get around the Fourth Amendment and say, well, that's not actually yours. It belongs to Amazon or whatever. We can go talk to Amazon. And Amazon has to stand in the middle of that process and say, we've invented another process to somewhat protect the people.

And I look at that, and when I was covering the first third-party doctrine cases that covered the cloud services and the government kept winning, that's basically when I turned into Joker. And I was like, all of this stuff that we're pretending about textualism and the plain meat. Like, none of this means anything because we just horsepowered a backdoor using this ancient law into everyone's data. And then I look at this, and I look at Anthropic, and I say, well, this is the same pattern, right? This is a private company saying, OK, we understand your position. We understand that you've reinterpreted the law to mean this thing. And we're going to put some process in between you, our tool, and the data of Americans flowing through our service.

I'm just wondering if you see that parallel there between Anthropic and Amazon and Azure and whatever other cloud services that exist that hold so much for it. Yeah, though I think there are a few clarifications that are important here that make this a little bit different. And in fact, you know, it was reported, I think the New York Times had this reporting first that, you know, the main clause that was most important to Anthropic was specifically about data collected from commercial services and not being able to use Claude on that data, which is... you know, exactly this issue in terms of third-party data. But I do want to clarify the main difference where what we were just talking about before this with, you know, Amazon or other third parties hosting your data, that was cases where they were, you know, because of where they sit in the ecosystem, they were hosting your data directly. With Claude, it's not that anyone is worried about the NSA looking through like your...

Claude usage, right? It's about them going out and getting third-party data from an Amazon or more likely, you know, the sort of sneaky hidden data brokers that serve ads on your phones and know all your location and your interests and things like that. And then feeding that into a system that Claude would then work on. That's what Anthropic really didn't want to be a part of. So wherever or however the government would collect that data from a third party, Claude said, we don't want our tool to be used on that data. There's a piece of that that just feels like Apple famously stands up to the FBI, put a backdoor on the iPhone, Apple says no, and they stand up to Trump. And there's just a part of...

However our system works in which big private companies get to say no to the government on behalf of their customers. And this felt the same. In the same way that Apple, again, won't put it back on the iPhone or the big cloud providers say there's a little bit of a process you have to jump through before you get the individual data. Here it seems like Anthropic is saying we're not just going to do bulk analysis of data that you have acquired from other parties because that leads to 24-7 mass surveillance of Americans and we don't want to do that. And that seems like a bridge too far for this administration.

Is there any coming back from that? I mean, we'll see. In the past when that's happened, and it's happened plenty of times with most of the large tech companies, at some point they've said something is a bridge too far. And where that normally goes is to court. And that, you know, the companies will go to court or the administration will go to court and there'll be some sort of court battle. I mean, you know, backdooring the iPhone is a perfect example of that. Sort of went to court and they sort of fought it out, but they never quite got to a conclusion because the FBI eventually did just...

you know, manually break into the iPhone and then didn't want the court ruling to ruin the future. But in this case, you know, where the escalation is and where this is different than those past situations is that rather than just going to court, they did this supply chain risk designation, which is just insane. You know, this idea that this tool, which was designed to stop, you know, potential foreign malicious actors from supplying technology that could then, you know, put in hidden surveillance tools into the larger technology stack, that those could be banned to apply that to a U.S.-based company. basically for having an ethics policy feels like a real misuse of that tool. And even that tool, I think, was questionable in some ways, but you could understand the impetus behind it for when you're talking about a Chinese networking firm or something along those lines. Here, it makes no sense. And so the reaction to this goes so far beyond what would normally be seen in this case. You could see, traditionally, there would be some sort of court case and either side could start it. And it would just be a battle about what...

how the contract could be applied. But that's not what is happening here. This administration is effectively saying, if you don't give us absolutely everything that we want, if you don't set up your tools to work the way we want them to work, then we will effectively try to destroy your entire business. And that is an escalation. There's one piece of this that I want to end on, and it's kind of the galaxy brain version of all this. FIRE, which is a free speech advocacy group, they put out a blog post just before I started recording, making the argument that forcing Anthropic to build tools it doesn't want to build is a free speech violation.

that it's something called compelled speech. And there's a lot of history here. This is some deep verge in tech dirt, like in the weeds, existential crisis history here. But it basically comes down to the idea that code is speech. Writing code for a computer is a form of speech and the government can't force you to do it. And a whole bunch of stuff flows from that. Do you buy this argument that forcing Anthropik to build tools that doesn't only build this compelled speech? Yeah, I actually think it is fairly compelling. Compelling, compelled speech. But no, I do think it is an interesting argument. It's one that had been a little bit further down the list of the issues that I was thinking about.

obviously mostly more focused on the Fourth Amendment issues and that. But this, I think the FHIR argument is not wrong. And we have seen this in other contexts in terms of, you know, it did come up in the backdoor issue as well, in terms of trying to build backdoors into encrypted systems. Companies definitely raised First Amendment claims and saying you can't, that is compelled speech to force us to write that kind of code. And so I think it is a valid argument. It might be.

Again, one that courts are probably less willing to tackle initially if they can deal with these issues some other way. But I'm glad that FIRE made that post, and I think it is an interesting and compelling argument. It's just the nature of the second Trump administration is that it's such a blunt instrument. It is almost certain we will attack all of the issues all around. Yes. Yeah, yeah. And, yeah, I mean, every Bill of Rights amendment has to be challenged in some form or another with every possible issue. I'm sure we can fit a Third Amendment violation somewhere in here as well. Oh, sure. Yeah, Claude has to live in your house. Exactly. It's going to be great.

We're doing one, three, four, and seven. Let's rack them up. Mike, this has been great. I cannot believe you haven't been on the show before. This has been great. You've got to come back soon. Absolutely. Whenever you want me. Before we wrap up, really quickly, I want to bring on a very special guest. It's Helen Havlock, our publisher here at The Verge. Hey, Helen. Thanks for having me on. Helen, I wanted to ask, because you run our business, there's a lot of talk about AI. We just did a whole episode about it. How are we investing in that coverage? How are we growing that coverage here at The Verge? You and I obviously spend a lot of time talking about The Verge's AI coverage.

how The Verge needs to cover the story. It's both the biggest story as a standalone beat for The Verge. We've hired some amazing people like Hayden Field who are covering as a standalone beat, but it's also the biggest story in kind of every desk of The Verge, right? Our policy desk is pointed dead ahead at the story and we have a lot of fantastic people there and Tina Nguyen who we've hired and Sarah Jong who's doing amazing reporting. And then lastly in our product coverage, you know, I think as you and I talk to each other and then as I go into market to talk to partners and advertisers about The Verge, I think what makes our coverage really unique is that it's rooted in the product coverage. So what our senior reviewers have to say about the actual products these companies make, how they are to use, what the experience of using product is as a consumer.

I don't think you can tell the business story of AI and the policy story of AI without being deep in the product story of AI. And I think that is what makes it. a perfect Verge story for this moment and what I think also makes The Verge the uniquely best place to cover the story in this moment. It is true that not enough AI coverage features the question, does this work? And The Verge relentlessly asks that question. Does this work and will people pay for it? We often describe Helen as our firewall here at The Verge. She runs the commercial side of our business. I run the newsroom. Helen runs the advertising business or subscriptions business. And there's a wall between that. Our business doesn't affect what we make here in the newsroom. And it's Helen and I that protect that dynamic. Helen is the firewall. Helen, help people understand that separation and what individual listeners and readers of The Verge can do to support us. One of the things that makes The Verge...

I think really unique and respected is how strong our ethics policy is, where our coverage is definitely not for sale. But we do still have a huge advertising business. That's a big part of what supports The Verge. And so a big part of my job is to work with our partners, our advertisers, our sales team, to help them understand The Verge, why they should advertise with us, what opportunities we can develop. You know, we've made new ad products like Quick Post ads. They're clearly disclosed.

We want to make good advertising that people like because that also makes ads perform better. And so in running that business, I'm kind of on that side of the house. We also, for the last year, I think as many people here know, have really big subscriptions business. And that's the single most important thing any individual person can do to support our coverage. in this moment is subscribe to The Verge. Benefits you get with a Verge subscription, you obviously get unlimited access to read The Verge. We now have an ad-free version of this podcast feed, Decoder, where you can subscribe and get the feed for free. And then if you are interested in the story of AI and policy and some of the themes of this episode, your subscription also gets you a bunch of newsletters you should really enjoy. Tina Nguyen has a fantastic newsletter about MAGA tech policy and how all of this is unfolding called Regulator. If you're interested in the product side of it, I'd really recommend Victoria Song's newsletter, Optimizer, about the crazy, like...

AI intersection with wellness and how we all live. David Pierce's Weekend Read Installer is always one of my favorite, like, what little things should I play with that are new and interesting? And there's a lot of product. playing out there. Right now, a Verge subscription annually, and because I'm the publisher in Firewall, I will shill shamelessly. A Verge subscription right now starts at $40 for the first year. It is incredibly good value, and it also supports all of this work we need to do and helps us invest in hiring more reporters to cover the story. I will tell you, I found out this week some of the biggest executives at some of the biggest companies in the world, Reed Installer, the same way you do every single weekend. It's pretty good stuff. One thing you and I have been talking about is the shift in the advertising market. Decoder listeners know this well. Influencer marketing is real. The idea that the ad should be integrated with the content is everywhere. You've got some ideas.

for how we can do that and preserve our ethics policy and keep that separation alive. We're going to hear some of those ideas from you on the show in a couple of weeks. Give the listeners a preview. Yes, I'm super excited to be piloting a new ad product for this podcast specifically. It's called Decoder Sessions. A joke we have that is also deeply true is that Disclosure is our brand. And so I think the way The Verge handles this is to make really good products that do a good job for our advertising partners, but that also ideally make a much better product for you, our listeners and viewers, while also disclosing it and separating it to the standards of our ethics policy. So one thing we are going to try on this show where...

The thing you cannot do is pay us to have your executive talk to Neelai. But this is a show of interviews. This is a show of interesting conversations about business and organizational structure and technology. So we are going to take that format, which is to interview interesting people from companies about their businesses and their organization. And we are going to make a new kind of ad break for you that hopefully you really enjoy and that is better than kind of.

a standard podcast ad that's more fun, more engaging, more native, I will host it. So I'm going to be talking to different executives from some of our partners. It will be separate from what Neely's doing. Neely will not get a heads up on any of it. But I'm really excited about it. What we want to make is a better product for you and a better product for our partners that still kind of holds the high standards of The Verge. So we have our first one coming up in a few weeks. I think it should be really fun. I'm talking to an executive from L'Oreal Group. It is going to be a really fun time. I'm excited for this product because I want to go compete for where I think the advertiser dollars are going, which, again, every decoder listener knows, I think, is headed towards the creator economy in ways that...

Collide with journalism. So this is our approach to solving it. That one's going to come out. You're going to hear Helen do that interview. We're very interested in your feedback and how we can make that stuff better. I think the ads should be good. I think they should be fun to listen to, just like the show is. I just think someone else should make them. So Helen, I'm excited for your first Decoder session. Thank you so much for joining me quickly at the end here. Thank you, Eli.

I'd like to thank Mike Masnick for taking time to join Decoder, and thank you for listening. I hope you enjoyed it. If you'd like to let us know what you thought about this episode or really anything else at all, drop us a line. You can email us at decoderattheverge.com. We really do read all the emails. You can also hit me up directly on Threads or Blue Sky, and we're on YouTube. You can watch full episodes at DecoderPod. We also have a TikTok and an Instagram. They're at DecoderPod as well. They're a lot of fun. If you like Decoder, please share it with your friends and subscribe wherever you get podcasts. Decoder is a production of the Verge and part of the Boxing Podcast Network. The show is produced by Kate Cox and Nick Statt. It's edited by Ursa Wright. Our editorial director is Kevin McShane. The Decoder Music is our Breakmaster Cylinder. We'll see you next time.

That's two minutes for holding. But you're holding two times the chicken with BOGO Box Meals from KFC. Order in and focus on the ice, not what's for dinner. Get game day deals on Uber Eats all season long, like BOGO Box Meals from KFC. Hockey is for the hungry. Uber Eats. Order now. Terms apply. Product availability varies by region. See app for details.

For most of the history of professional sports, there was a wall between the game and gambling. Then suddenly it was everywhere. Ontario is all in on sports. But that wasn't supposed to include the players themselves. Athletes are the perfect gamblers. Understood. The big bet. Available now on Spotify. Should you get a dog? A great dame? A teacup chihuahua? A labradoodle? Golden doodle? A pool noodle? Maybe just a cat? Whatever you choose, it's up to you. Just like it's up to you to ask your doctor about Wigovi.

Frequently asked questions

How long is this episode of Decoder with Nilay Patel?

This episode is 48 minutes long.

When was this Decoder with Nilay Patel episode published?

This episode was published on March 12, 2026.

Is there a transcript available for this episode?

Yes, a full transcript is available for this episode. You can read the complete transcript on the episode page.

Can I download this Decoder with Nilay Patel episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.