ICE DNA Collection, Google Earth AI Slop, and White House AI Cybersecurity

Zoë Schiffer: Welcome to WIRED's Uncanny Valley. I'm Zoë Schiffer, contributing editor.

Brian Barrett: I'm Brian Barrett, executive editor.

Leah Feiger: And I'm Leah Feiger, director of politics and science.

Brian Barrett: Today on the show, we're talking about how the Immigration and Customs Enforcement Agency, better known as ICE, has been collecting DNA samples at an alarming rate. These samples have led to hundreds of thousands of people, including young children, to end up in an FBI criminal database basically forever.

Zoë Schiffer: We'll also get into one of our favorite topics, AI slop. From Google Earth disabling its new AI tool after people used it to create fake maps to Substack partnering with the AI writing detection tool Pangram, companies are starting to realize that people really don't like low-quality AI content flooding their feeds.

Leah Feiger: Also in the realm of AI, the White House announced a plan to tackle the increasing security risks brought on by AI companies. It's keeping the details mostly under wraps, but we're going to discuss what's behind all the secrecy and what this could mean for future regulation.

Brian Barrett: We'll also talk about how part of a SpaceX rocket crashed into the moon, leaving a new crater behind and all kinds of questions.

Leah Feiger: And later in the show, WIRED senior writer Molly Taft joins to tell us all about how data centers have officially broken US politics.

Brian Barrett: Well, to start off, let's talk about ICE, which we haven't in a little bit, but they're still there. They're still doing bad things, including collecting DNA samples. WIRED’s Dhruv Mehrotra reported this week about how ICE has collected nearly 1 million people's DNA this past year, just this past year alone, including that of young children. As we mentioned up top, this DNA collection has led to hundreds of thousands of people with no criminal convictions ending up in an FBI database indefinitely. And in fact, the Department of Homeland Security has become the single largest source of new profiles in this database.

Leah Feiger: I feel like we're getting inundated with here's one more part of your privacy, here's one more part of security, here's one more bit of immigrants being treated as subhuman. But this one, at least enough for me, broke through the noise. Don't you guys think? This was wild.

Zoë Schiffer: I feel like I have a question before we move forward, which is maybe one that's shared by listeners. Why is DNA being collected from immigrants in the first place? This surprised me, and maybe I'm a little naïve about what they're subjected to when they're in this country.

Brian Barrett: Yeah, I mean, I think the answer is because they can. And I wish that we're more nuanced or subtle than that, but I think they are just trying to gather as much information as they can at all times. I want to be clear too, these are not people with criminal convictions, people who are even necessarily suspected of a crime. It is just such a wide net that they are casting. In fact, internal training materials that drove, in partnership with researchers at Georgetown, we should say, dug up, said that asylum seekers and refugees who have not adjusted their immigration status must have their DNA sampled and that the DNA needs to be collected after any arrest. It is just table stakes for these officers to grab DNA whenever and wherever they can.

Leah Feiger: And already a lot of lawmakers have been speaking out against this. For example, when learning that children were being swabbed at a family detention center in Dilley, Texas, a number of representatives said in a joint statement to WIRED, none of the families at Dilley have been convicted of a crime. They do not belong in a database meant for violent criminals, especially children. This is starting to pick up steam across the board. I think that we've just gotten so used to the idea that the government is allowed to collect any information it wants, store it in whichever way it wants. And all of these norms have been destroyed in the last year and a half. It's really shocking, honestly.

Zoë Schiffer: So you're saying basically that refusing to allow your DNA to be collected can become its own chart, can become a mark on your record regardless of what you had previously.

Brian Barrett: Not only a mark on your record. WIRED and Dhruv identified two prosecutions filed in 2025 against people in immigration custody who refused DNA collection. It is becoming a crime to say, "I don't want you to do that." That's one thing. Then the other thing too that I think we've mentioned, but I want to dig into a little bit, is that they have started recently... I think the shift to collecting DNA from families is also relatively new. And we're talking about, I think the official line is that kids as young as 14 can have their DNA collected, but Georgetown and WIRED identified 492 children under 14 whose DNA the agency sent to the FBI, not just collected, but sent onto the FBI between January 2025 and January 2026. That included 21 5-year-olds, 32 6-year-olds, 33 7-year-olds who are now in an FBI DNA database for the rest of their lives.

Leah Feiger: I mean, the kicker to me here too, part of this has to do with "preventing future crime." It's all supposed to be about how the primary purpose is to build a national DNA database for solving crimes, which is a theory. OK, sure. Solving crimes, cool about that. But these are 5-year-olds. These are 5-year-olds whose—so many of their rights have already been taken away by this current administration and DOJ and DHS. To have that as an additional marker to me is a really important visual next step of the war on immigrants at this moment in the country.

Zoë Schiffer: All right. I'm going to take us from the depths of despair. Just drag us up.

Brian Barrett: Some would say AI slop is maybe not the depths, but we're definitely still in despair territory.

Leah Feiger: I mean, I love the depths of despair, my favorite place to be.

Zoë Schiffer: I find it funny, but I have a bunch of thoughts on it and my views on this have been evolving really rapidly, so I'm excited to talk to you about it.

Leah Feiger: Good, good, good, good.

Zoë Schiffer: Last week, Google rolled out a new AI feature on Google Earth, which would basically let anyone zoom into real coordinates on a map and type a prompt just like you would in ChatGPT or Gemini and superimpose an AI scene on top of real Google satellite imagery.

Brian Barrett: What could go wrong?

Zoë Schiffer: Yeah, I have a hard time imagining how no one during this process said, "This is a terrible idea," or perhaps they did and they were not listened to. But regardless, Google did this. As you might imagine, it took 24 hours for people to generate fabricated nuclear plants in Iran, a bombed hospital in Gaza, and fires at an Iranian oil terminal. People were extremely upset about this for misinformation, disinformation, and other reasons, and Google quickly disabled the feature. It just feels like there is a real split between how Silicon Valley feels about AI and uses AI and how the rest of the country feels about it and uses it, because these tools are pretty unpopular, even as they're being widely adopted. But I think that the way that they're being adopted is also really different. Most people are using ChatGPT, Gemini, Claude as basically a replacement for Google. Whereas in Silicon Valley, they're using it as a replacement for a personal assistant or a junior employee. And so for them, I think there is real magic in the tools and the capabilities. But for everyone else, the thing you hear a lot is why is this being shoved down our throats? We don't want this. They're stuffing AI features in all of these apps that we previously used and were good, and the AI features are bad.

Brian Barrett: Well, I think it's AI slop specifically, right? Yeah, there are use cases for AI agents. Zoë, you've given over some percentage of your life to them already. It's working out great for you. But what I find interesting about this is (A) how quickly they rolled it back, and (B) it's just the latest in a series of steps like this where companies that have been sort of embracing AI up to this point are acknowledging through actual feature changes to their products that, hey, maybe people don't like this. Just in the last week or so, you had LinkedIn adding a little button to a post where you can say, "I think this looks like AI slop. I don't like this." LinkedIn, which has basically been a platform that has been built the last year or so on saying, "Hey, look at all the AI slop you can make on this platform." That's their whole thing.

Zoë Schiffer: Love AI slop.

Brian Barrett: But now they're saying, OK, we can actually give people a chance to say no. You had a couple of weeks ago, Substack, in the app, you can check to see if it's potentially AI generated. They partnered with a company called Pangram, which is a very good AI detector tool. Again, Substack, a company that has said in the past, "Hey, if people want to use AI, that's great." And then more broadly outside of companies themselves, if I can monologue slightly longer because I'm passionate about this.

Zoë Schiffer: No, please go. Go off.

Brian Barrett: The EU on Sunday, a new regulation kicked in that said, "Look, anytime you come across AI slop in your feed, whatever it is, it has to be labeled as AI slop. Every piece of AI content has to be labeled as such," which is a huge sea change for people over there. If you're an AI creator, if you're just someone on social media, it's telling that, yeah, people want to know when they're seeing it and they want the option not to see it. And companies are listening.

Zoë Schiffer: It's interesting because I was hearing this a lot from actual AI researchers at OpenAI, Anthropic and Google, and they were saying it's so annoying how all of our executives are constantly just posting their AI. They won't just post their own messages on Slack. It sounds very distinctive. It's not that it's posting misinformation or getting facts wrong. I think largely the tools really have improved on that front. You need to check them. You need to be literate on that, but it's more that the patterns and the way that they speak is distinctive and there's an ick factor for people. I really am noticing that these days, the difference between an executive who is a good writer and actually saying something that they're passionate about and writing it themselves and posting it on social media, and then someone who very clearly, whatever, maybe they had a conversation with Claude and it's just 100 percent AI-generated output, my eyes glaze over it. It is genuinely hard to read this stuff. It's hard to remember what you just read. I think is getting bigger and bigger. And I think that real human writing, real human output and content is going to be an edge for companies that are willing to invest in it, which is not to say that there isn't cool uses of AI. I actually had a real unlock this week, because nothing unites writers and reporters more than how much we hate our CMSes. I legit just built a better one. It works beautifully. You can post blogs. It took me no time at all. It really was like a, OK, wow, these tools can do amazing things. But that's very, very, very different than what we're talking about with Substack or Google.

Brian Barrett: I want to take a quick moment to flag that Zoë just used unlock as a noun.

Zoë Schiffer: Oh, no.

Leah Feiger: It was painful. Just to be clear, I'm with Brian on this.

Brian Barrett: Yeah. I felt like it was important to say that. And then we'll move on. And then we'll move on.

Zoë Schiffer: I deserve that. You spend too much time in Silicon Valley and you become one.

Leah Feiger: OK. I am game to keep talking about AI, but let's pivot a little bit to the White House, if that's all right with you guys.

Brian Barrett: It is.

Leah Feiger: Great. Great. I feel like we've talked about this, honestly, a bunch, which is the White House is trying to figure out how to regulate AI, specifically keeping in mind cybersecurity risks. We talked about how the Trump administration has had different officials weighing in, how people have had some fun little disagreements. Apparently, as of this week, they have now all arrived at a common framework, and for the most part, they refuse to tell you anything about it.

Zoë Schiffer: Love that.

Leah Feiger: Love that. On Tuesday, the White House brought together staff from OpenAI, Anthropic, Google, Meta, Nvidia, and a couple other AI companies to walk through its finalized AI cybersecurity oversight plan. The details of the plan haven't been released publicly, but we do know that developers can voluntarily submit new models up to 30 days ahead of release, and then the government is going to test their hacking capabilities using a classified benchmark. From there, they're going to share the results with federal agencies and "trusted corporate partners." Reportedly, open weight models are being entirely excluded. I feel like I said a lot of details, but honestly, we're not getting a lot of details yet, really, about what's included here. A lot of companies feel like they're being left out. The White House has had some wild quotes about how they do have to keep things classified and even