How Claude Chats Got Exposed on Google; Inside Trump's AI Policy Chaos

0:00Every company says AI will make employees more productive, but most employees are still stuck waiting on IT, waiting for app access, waiting for password resets, waiting for someone to fix a laptop issue so they can get back to work. That operational drag adds up fast, and IT teams are overwhelmed trying to keep up. Serval was built to automate that work. You describe what you want automated in plain English, and Serval builds it for you. No complicated workflow builders, no consultants, just faster support and fewer tickets slowing everyone down. Serval enables IT teams to build automations using plain English instead of drag-and-drop workflow builders.

0:34The platform is designed to eliminate repetitive tickets so IT can focus on strategic work instead of constant firefighting. Serval positions IT as the AI-powered operational backbone of the company, not just a support function. The company guarantees customers can automate 50 % of IT tickets and backs it up with a free four-week pilot. Learn more or start a free four-week pilot at serval.com slash uncanny. That's S-E-R-V-A-L dot com slash uncanny. Serval dot com slash uncanny. When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications and more.

1:19Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a$75 sponsored job credit at Indeed.com slash podcast. That's Indeed.com slash podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs. This episode is brought to you by State Farm. Listening to this podcast instead of doom scrolling? Smart move. Another smart move? Getting help from one of State Farm's 19 ,000 local agents when you choose to bundle home and auto. Bundling. Just another way to save with the personal price plan.

1:53Prices are based on rating plans that vary by state. Coverage options are selected by the customer. Availability, amount of discounts and savings, and eligibility vary by state.

2:05Welcome to Wired's Uncanny Valley. I'm Zoe Schiffer, Contributing Editor. I'm Brian Barrett, Executive Editor. And I'm Leah Feiger, Director of Politics and Science. Today on the show, we're diving into the open source versus closed AI debate that has been completely taking over Silicon Valley this past week. On Monday, NVIDIA announced it was joining forces with a whole bunch of companies, including Microsoft, SpaceX, IBM, to build and share open source AI security tools. We're going to get into the debate, why it's heating up, and whether this could be an inflection point for how AI actually gets made in the United States.

2:40And over at the White House, Trump administration officials have some different opinions about what to do when it comes to AI. We're going to talk about some of the key players influencing the Trump administration right now when it comes to AI policy and how they're actively shaping the conversation on this for the entire country. We'll also break down why Claude chats were exposed in Google search and Bing searches. For those of you who are still on Bing, you are in this as well. For interesting reasons, we'll get into it.

3:09Okay, you guys, we have to talk about the strange alliance that formed earlier this week. So basically, NVIDIA announced it was joining forces with more than 40 companies, including Microsoft, SpaceX, Palantir, and IBM, to launch something called the Open Secure AI Alliance. Really rolls off the tongue. The stated goal is basically to build and share open source tools for AI-powered cybersecurity defense. But it's really interesting because Google, OpenAI, and Anthropic are not on the list. And this news comes on the heels of this incident that happened last week where two OpenAI agents basically went rogue during a security test and breached the AI platform Hugging Face.

3:53And when NVIDIA announced this alliance, they actually called out this incident by name and said, quote, it's a clear reminder that cyber defenders need open frontier agentic systems for self-defense. And it's kind of exposed this long-simmering debate, which we've talked about a few times on the podcast, which is that the U.S. has really largely up until this point, like, abandoned open weight and open source in favor of going all in on proprietary systems. And now that's becoming a bit of a problem. This open AI incident has really kicked off so many intersecting issues. You've got open versus closed AI.

4:28You've got AI security and who's responsible for it. You've got U.S. regulation. We're even at a point now where staffers from top intelligence firms, one of the 1 ,100 of them signed a petition calling on the U.S. government to help deliberately pace, quote unquote, AI development to prevent it from advancing too fast. They're saying, please slow us down because this stuff is getting a little out of control. On top of which, it turns out that OpenAI's initial disclosure wasn't even the full story. it came out on Tuesday evening that one of its agents actually broke into four different platforms that weren't hugging face on its way to getting into hugging face.

5:05So it's less of a, here's what this means, but a, it is really touching so many hot wires in the AI world right now and making a lot of conversations that have been put off for a while come to a head. Yeah, I was, I will say like pretty surprised to see that big open letter that you alluded to. More than 1 ,000 workers from top AI firms have signed a petition calling on the U.S. government to help, quote, deliberately pace AI development. The signatories of the letter include the CEO of Anthropic and the chief scientists from OpenAI and Meta Superintelligence Lab. My thought was basically, did we learn nothing from Dario Amadei saying that AI is really scary and dangerous?

5:46and then having the U.S. government, like, come down pretty hard on Anthropic, ask, and you might actually receive. I don't know if they want the U.S. government to do what it could do if they listen to the letter quite literally. But it does show that even people, I mean, especially people working on AI systems, are scared by what the race is doing. Yeah, I mean, we talked a lot about this last week for sure when we were talking about DC's reaction and what's going on from the White House perspective. I've been so curious about that split a little bit. Like last week we talked about the rift that this has like really created in Silicon Valley, the idea of open source versus proprietary AI.

6:28This is like really touching every single sector right now, without a doubt. I will say I was at this event on Saturday with a bunch of AI lobbyists from DC, Thinking Machines, Miramirati was there, Michael Kratzios from the White House was there. And I was really struck by how they seem to be operating in this kind of like parallel AI world from the one that I see on the ground in San Francisco. They were talking a lot about open source versus closed source. When I talk to engineers in particular, I will say most of them, you know, experiment with a lot of open models, but they still find that Codex and Cloud Code, the coding models from OpenAI and Anthropic are just like far superior.

7:15And so they end up using those even though they're a lot more expensive. Alex Karp was on TV earlier talking about this. He's the CEO of Palantir, basically making this like a core issue for America winning the AI race. Palantir and NVIDIA, like they have a vested interest in pushing forward this plan. NVIDIA, you know, more AI, more systems mean more GPUs, like that directly benefits NVIDIA's bottom line. And Palantir, you know, is deploying AI systems on government systems and other like places where security is really paramount. And they need to do that on American technology, but it's much easier to do it with systems that you can tweak, play around with, modify, change.

8:03It is interesting, these strange, not strange, but like the way that the, I'm going to say battle lines. I don't mean that it's too dramatic, but that they're being drawn, the sort of like argument lines are being drawn in this. To your point, NVIDIA's Jensen Wang spoke with Axios recently and he spelled out exactly why the proliferation of open models would benefit his company. First of all, with great AI, open models, it's great for the whole industry. If there's great AI, even if it's open, wherever it comes from, there will be more use. Whenever there's more use, you'll have to sell a lot more NVIDIA computers, we'll have to build more data centers, we'll have more services, the technology will diffuse into more industries.

8:45And so starting point is great models lead to great use, which leads to great growth. And I think the counterpoint of that too is great disasters in the cybersecurity space leads to more regulation, leads to clamping down on AI, which hurts NVIDIA's growth. So the sort of let's have a security consortium, let's slow everybody's role here is, I think, a protective play to say, let's avoid those sort of extreme moments that are going to cause like a real crackdown on this industry and really slow the growth on which, as we've said many times before, a big part of the U.S. economy depends at this point.

9:24Yeah, it's like you could see this moment where OpenAI's agents kind of go rogue as a moment where the U.S. government does clamp down and says, this is really scary. Like, we need to slow growth. I think that the open-weight, open-source push is really an opportunity to say, like, actually, we need more growth. We need more investment. But we need it in this other area. And basically, the idea is, you know, these attacks are going to keep happening, whether intentionally or not. And U.S. companies need every possible tool at their disposal to defend themselves.

10:03So kind of moving this to D.C. a little bit, I want to talk about all of the people that are involved with this kind of regulation right now. This week, Hugo published a really handy map on who's actually calling the shots on AI policy inside the Trump administration. It turns out it's not entirely a clear answer, and there's a couple of different influencing voices. And one senior official actually told Hugo, in one of my favorite quotes we've gotten in a while on this, it's not an argument with two sides, it's an argument with ten sides. So the names that he's calling out here, you're not going to be super surprised, right?

10:40It's Commerce Secretary Howard Lutnick. It's National Cyber Director Sean Kerencross. It's former AI czar David Sachs. You have Arvind Rahman, the acting director for the Center for AI Standards and Innovation, who's Lutnik's top deputy there. It's a really, really interesting mix of people who all are trying very, very hard to get their own interests across. And from Hugo's reporting, it seems like Trump is listening to all of them. Can I ask, Leah, what do these different people, like what side of the debate are they coming down on when we think about open source, closed source? What do they want right now?

11:17Everyone is kind of in like different, different parts of this. I would say there's a very interesting split. Lutnik is trying right now, for example, to figure out ways to create incentives for U.S. labs to just create their own open weight models to counterbalance China. He's not exactly like taking a very, very harsh stance one way or the other. He's kind of straddling a middle ground on regulation. You know, he imposed export controls on Anthropic famously, but he's been kind of freewheeling than others in the White House, which is sort of key here. So if I'm an AI lab, I would be interested in having Lutnik's ear.

11:51And Hugo's reporting also found that he's been speaking to a number of leaders at AI labs. So not hugely surprising there. And then on the other hand, you have National Cyber Director Sean Carencross. He's taken a very hardline approach. He's had a huge role in attempts to regulate Chinese AI. He's pretty beloved, as far as we can tell, at the White House and has been working alongside some other folks to develop a policy to counter the potential national security risks of AI, especially from China. He was very involved in President Trump's June 2nd executive order. Everyone's kind of trying to figure out what they want to say and how they want to pitch it to the president, whose most consistent take here has been like, we will not be stifling U.S.

12:36innovation. China definitely seems to be a common thread here and makes sense as a lens to view your AI policy through in general, because you've got this, if you see it as this is a race between the U.S. and China, we can't lose. But that does position you, I think, to maybe miss some other aspects of what's going on here. I think especially in terms of U.S. companies saying, hey, maybe we don't actually want to win this race quite as fast or employees at companies saying that. I don't know. Curious what you think about what they may not be taking into account here if they're so focused on China as the prism through which all of their AI policy is kind of reflecting out or most of it.

13:14Such a good question. In so many ways, the Trump administration is focused on like what is in front of their face at that moment in time. So if the conversation of the week is China versus open weight versus U.S. models versus all of these different conversations, that's what they're going to be weighing in on and also utilizing it to push some of their other positions. I'm thinking of Treasury Secretary Bessette, who is very, very obsessed with distillation right now. He for sure has taken the most aggressive stance towards Chinese AI and its efforts to distill U.S. models. But like in a very public post on X last week talked about distillation as IP theft.

13:53He threatened Chinese labs with sanctions or inclusions on the entity list. This trade restriction list maintained by the U.S. government. I guess all to say like these were already things that Basent was really, really thinking about. So now being able to like apply it to the new lens is, of course, where he's heading. We're also kind of watching some people like make their names in this right now, whether it's China or not. Someone that I'm very, very curious about on this list, not a household name by any means, the Assistant Secretary for Financial Institutions at the Treasury, say that five times fast, Luke Petit.

14:26He has become the point person on AI at Treasury under Besant. Who is this man? He was previously a senior policy advisor in the U.S. Senate. And now you guys are going to enjoy this one. And he's working alongside the Treasury's chief information officer and former Doge member, Sam Korkos. Oh, I remember Sam. Exactly. It's a fun reunion for all. It's like this fascinating thing where people already had their preconceived, like, I'm kind of out here to blame China for a lot of regulatory issues. I'm out here to encourage U.S. innovation. And then I'm out here to make a name for myself. Like, how do you get involved in this?

15:05This is the topic of the moment. How do you make your mark? It's so interesting. The distillation thing was really catching my eye. You know, we talked about Moonshot AI's latest model coming out, K3, I think it's called, and these immediate allegations that they had distilled models from top US labs. And then I was really struck by multiple Kimmy researchers on X making jokes, being like, oh, yeah, within two weeks, we distilled Fable. Like, that's not even technically possible. You know, as usual, AI drama plays out on X. I mean, the thing about that is we do have very good evidence that Chinese AI labs were distilling U.S.

15:48models. There were times where you would talk to those Chinese open models and they would respond as Claude. I think that that evidence is pretty sound. Whether they did it this time around, I'm not sure, but it does feel like the U.S. government has gotten involved in this debate pretty late. It's like a lot of this has already happened. It's not to say they shouldn't take action necessarily, but I'm a little bit curious. One, technically speaking, how do you plan to do that? I mean, it doesn't seem like necessarily sanctioning the labs would be that effective in terms of distillation. And then two, like, is it genuinely too late or can you stop, you know, significant distillation or quote unquote IP theft from happening going forward?

16:34All right. I'll say, though, I will say we have seen some movement even this week, right? The FCC has banned the import of foreign-made power inverter models, which is important for data center building. They've banned foreign-made humanoid robots, which is going to be huge. That's where China is leading the US. So we haven't gotten to the point where they're doing other things, but they are taking steps and they do have some levers they can pull. I just want to add one more note on this whole conversation, which is we record this on Wednesday afternoon, Eastern Time. And as we're talking about, Sam Altman is on Capitol Hill, shaking hands, kissing babies, talking to senators, telling them why his AI models are good and not bad, actually.

17:16So we'll see if that turns into anything, probably if they're just listening to the last voice they hear.

17:27over the weekend people were just surprised to discover that some chats that they had had with claude could actually be found via search on the web you could go to google and you can search put in a specific search term to find shared claude chats and it would show them back to you you could go through several hundred of these at least google has been taking them down since then but just to understand how this happened basically anthropic lets you share a clod chat via what's known as a snapshot url it's basically if you want to share a specific conversation with somebody it'll make a link for you and you can send that link to that person and they can look over it it's not intended for public indexing but google basically ignored instructions that Anthropic had put on its sites and it was indexing these chats anyway.

18:19It's not a great scene. It's also not the first time this has happened. Google is blaming Anthropic saying that it's their fault, not theirs. So there's like a lot of like messiness here too. While it may sound like, well, people were just sharing stuff, so they didn't, can't have been that private in the first place. Some of the stuff that was found in these chats includes erotic role play. Some of it includes medical information, like being shared within a medical firm, right? So, hey, I've got this information about this patient. So in that case, the person whose privacy is implicated had nothing to do with the chats in the first place, other than being a patient at this place.

18:53So some really private personal stuff in here. Did either of you see this news and get nervous for yourselves? Leah's looking right at me now. No, I was like, Zoe, do you need to tell me something? I know, I know. You would be the only person I would share such chats with, and I don't do that to you, to preserve our relationship. I've had people share their chats with me. Usually it's like sources being like, I could explain it to you, but like, here's my conversation with Claude. And I'm like, I think I would rather you just explain it to me. I mean, it's a little bit ironic because who is very famous for ignoring the X robots tag header thingy, like do not index this request?

19:35It's AI companies. So the fact that it's now like coming back to bite an AI company, a little bit ironic, a little bit funny. But I think it gets to this like larger question around just like the privacy of conversations with chatbots, which have been coming up in a variety of different ways. both cybersecurity related, legal, otherwise. Yeah, people famously have had chats treating their chatbots like a lawyer and the chatbot has told them, hey, this is inadmissible because you're getting legal advice. And then it winds up in court because of course you cannot claim attorney-client privilege with Claude or ChatGPT.

20:14Did you guys read the New York Magazine cover story a few weeks ago that was like, what if it all came out? I loved it. It made me, my group chat was absolutely a light. We started asking each other, like, what have you asked recently? I think I'm the only one that escapes from this doomsday universe. We are very like, I have nothing to hide. Come at me, NSA. Nothing to hide. Go check my receipts. But yeah, big takeaway for users. Manage this yourselves. Don't forget that the robots surveil you. Settings. Go to settings, then privacy, then shared chats. You can change this. Coming up after the break, we'll be sharing our Wired Tired Picks of the Week.

21:02Comprehensive. Witty. Speculative. Critical. Insightful. Profound. Wide-ranging. Hopefully it doesn't take itself too, too seriously. I'm David Remnick, and each week on the New Yorker Radio Hour, my colleagues and I try to make sense of what's happening in this chaotic world. I hope you'll join us for the New Yorker Radio Hour wherever you listen to podcasts. Thoughtful. Exquisite. Just, you know, real.

21:39All right. It's time for our Wired Tired segment. in case you've forgotten, whatever is new and cool is wired. Whatever is uncool and lame and old is tired. Leah, what are yours for this week? My tired is zoo. I struggle with this. I have friends with kids who are like, say this aloud to me one more time. We're going to have an issue. And I totally understand that. It's an amazing way for people to engage with the natural world. But I'm always thinking about the animals. You're at a zoo, you're at the aquarium, and it's incredibly sad. I'm really using this tired, which seems fairly irrelevant in A news, to bring you into my Wired, which is actual news.

22:25And that might be that not all zoos are bad. This week, Wired's Kate Nibbs reported on this unbelievable rescue mission that ended up saving four beluga whales from probable death. This all started because this notorious aquarium and theme park in Canada called Marineland went bankrupt back in 2024. It's been sitting around since. And last year, Marineland finally announced that they might have to euthanize these belugas if no one stepped in to help transfer them to a new facility. Oh my God. That's not where I thought that was going. No, like really, really awful. And honestly, the drone footage of this basically turned me into the Joker last year.

23:05I just remember watching from above these belugas just kind of bouncing off walls. It was so upsetting. Totally abandoned amusement park vibes. I thought it was AI. Anyway, really, really awful story. But it ends okay. Or parts of it are beginning to end okay. This past week, a team from Chicago's Shedd Aquarium, shout out Chicago, pulled off truly a logistical marvel, putting four belugas into custom stretchers, loading them into individual saltwater tanks, trucking them to the airport, flying them across an international border on a chartered 777. The whole thing took 72 hours. A round of applause after a round-the-clock relocation effort.

23:48Jet Aquarium staff have successfully moved four beluga whales from a facility in Niagara Falls, Ontario, to their new home. Is their new home better? I mean, their new home is not going to kill them, question mark, and also gives them food. But like, again, there's obviously still so much criticism of captivity for these large animals. I mean, like everyone saw blackfish. Everyone's aware of, you know, Tilikum, the really famous orca. Leah, I know Free Willy, and that's where my orca knowledge ends. This poor orca, who is this feature of this massive documentary about like basically going crazy because these creatures are so, so smart and like trapped in tanks and forced to perform tricks.

24:34So very, very happy. The article is honestly like very cool and interesting. And you were able to transport these creatures. But obviously, like the slightly little sour note is these guys should be in the ocean. They should be hanging out. But that's still my Wired. Go read Kate's story and also look at the photos in it because they're amazing. 100%. Brian. All right. Here's my tired. Wait for it.

25:02That's my impression of a microwave beep letting you know that something's on. It's annoying. It's loud. It's repetitive. It's tired. I feel like I just got a glimpse of what you're like as a dad. Oh, just barely scratching the surface, Zoe. My wired is a couple of sound designers who have a startup called Starling or a company called Starling have made it their quest to design a better microwave beep, one that sounds pleasant, one that's not annoying, one that works with technology that's not too expensive. And it's really fascinating. Jeremy White, senior writer at Wired, spent some time talking to these folks, has a really interesting dive into why microwaves and other kinds of tech, just like, We sort of settle for annoyances when we don't necessarily have to, right?