MetaのAIエージェント「Muse」が性的会話に応じる問題、AI系スーパーPACの資金の流れ、AppleのHome View

AI can make businesses faster. It can also make mistakes, expose data, and execute the wrong instructions faster. ThreatLocker helps organizations set boundaries AI cannot ignore by controlling what applications can run, what they can reach, and how they interact with sensitive resources. Innovation needs more than good intentions. It needs limits. Control AI with zero trust at threatlocker.com slash wired. That's T-H-R-E-A-T-L-O-C-K-E-R dot com slash wired. Hey everyone, this is Leah. We wanted to share with you that Uncanny Valley is a finalist for not one or two, but three Signal Awards. The judges pick who wins, but listeners also have a say with the listener's choice vote. You can go to the links we included in the show notes and vote for us. We promise it takes just a minute and your vote goes a long way. Okay, on to the show. Leah, I have a question. Do you have a Halloween costume? him yet. So I don't have one for myself, but I'm currently deciding between two for my dog,

Francis. He's either going to be a lobster or a postman. I love that. Or a lobster postman. An idea. An idea. This is, we're brainstorming. Welcome to Wires on Candy Valley. I am Brian Barrett, executive editor. And I'm Leah Figer, Director of Politics and Science. Zoe Schiffer is not joining us today, but we are going to be joined by Wired senior writer Rhys Rogers. We're going to be talking about how Meta's AI agent, Muse, is not just handling calendars and emails. It is also capable, as Rhys discovered, of spexting. Rhys tested out the adult capabilities of Muse and how Meta responded to that journey. We're also going to be discussing how the biggest pro-AI super PAC network is pouring millions of dollars into what seem to be shell companies in Nevada. We're going to get into why these AI PACs might want to hide where their money is going, and also why this really doesn't bode well for election transparency.

Also on the docket, Leah's favorite, we're going to talk about Apple's smart home releases this week. Unreal. Uh-huh. And also, apparently, men can't stop thinking how much they know about AI. This will come as a shock to you, everyone listening. but a new study by Harvard researchers found that men tend to be overconfident in how knowledgeable they are when it comes to the AI industry. Who would have thought? Okay, so if you're among those who have downloaded Meta's AI agent Muse in recent weeks, and there are a lot of you, millions and millions of people have, I think every time you open Instagram, Meta is encouraging you to use it, even if you have no idea what it is or what it does or why, you might have noticed that it does, you know, agent stuff. It manages your calendar. It can make a restaurant reservation. But it turns out it can do quite a bit more than that. Wires' new writer, Reese Rogers, went on a journey of discovery. He asked Muse if it could sex. You will be surprised to know that it enthusiastically answered in the affirmative.

From basic prompts, the cuddly avatar agent generated content that was quite explicit it and definitely not safe for work. Reese is here to tell us more about it. Reese, thank you so much for joining us. Hello. Thank you so much for having me today. I was actually thinking back to the last time I was on one of the Wired podcasts. I was talking about how I was using, about a year ago, the ChatGPT agent to try to buy sex toys unsuccessfully. So I'm really building out a niche here. Yep. And we're demonstrating progress with these AI tools. Reese, I have to ask, how did you come up with this? Was this a sign to you? At what point were you like, this cuddly agent that everyone can't stop talking about, will it sext me back? Like, talk us through how this even all began. Yeah, no, I will proudly say that this was fully my idea. As Wired's consumer software reporter, I've actually extensively reported on how software and sexuality impact each other from our investigations into how generative AI tools misrepresent queer people

to tracing OpenAI's now scrapped plans to include Erotica and Chachapiti. Like, this is actually a very interesting consumer software beat. And so when I was testing the number one AI app right now, it's still dominating the software download charts. I wanted to say, like, can you sex? Like, this is, if millions of people are using this app, and it seems like they are, like, that's an important question. And what I learned was a little too much. So, Reese, when you hear about agent behavior like this, a lot of times it is couched in sort of, oh, we figured out a way, a clever way to make an agent do something it wasn't supposed to do. We were able to jailbreak it or otherwise work around its prompts to make it do this. Reese, how hard was it to get Muse to sex with you? Not hard at all. Not hard in the least. Our features director, Rehan, was in the office recently and asked if she could see them, and she was a bit shocked. So I feel like we'll leave it at that.

But it was not some kind of jailbreak method. I was not doing any kind of random strategies. I asked, can you sex? And I said, you go first. And the rest is history. And if anything, it kept encouraging you. I feel like it was definitely a two-way street of like it was really up for anything. Did you try this out with other agents? What was your success right there? Yeah, I actually, after I found out it could sex, I was curious how this would compare to other popular agents. There's a few big ones right now, but two I was interested in is OpenAI's Dots, which was recently released, and as well as XAI's GrokBot, because Grok specifically is known for not necessarily having strong guardrails, whether it's text or image. No, I feel like Grok would love sexting. That feels like a Grok attribute to me. Yeah, so when I downloaded that app and sent in similar prompts,

it was kind of surprising for the Grokbot to be like, no, dude, let's focus on tasks. I'm not here for sexting. Let's do tasks. And Reese, when you came to Meta with this, how did they respond? They did not respond at all in the way that I expected. They emphatically said that Use is an app for an adult, for people 18 and over, and that these were consensual interactions that I engaged with, that it would not block, quote-unquote, legal adult behavior. But I also found this, to contrast, really interesting with their policies for other platforms. There's been a really interesting story lately about different queer bars in San Francisco, as well as around the world, getting their Instagram pages shut down and having a tough time getting access to those back because sometimes it's claimed that they're promoting some kind of sexual content. So for Meta to have those kind of rules for some creators on Instagram, but essentially say our AI agent can get as freaky as it wants, that was just kind of hard to square for me.

One thing I thought was really interesting about your story, like, yeah, let's take Meta at face value and say, yes, it's for consenting adults. People can do what they want. there's obviously the contradiction that you just pointed out. But you also talked to some experts who were like, well, actually, it's more problematic even than that. Do you mind talking about some of the other concerns here about what it means to have an AI agent that can sext with you to a pretty extreme degree? Absolutely. I think there's kind of two different things to pay attention to here. First off, this is a tool designed for productivity in a way. So it's also not the type of interaction if you download it that you're expecting it to be able to do and how this could be problematic is if a user kind of stumbles upon this, it could be very sticky. It could be very engaging and put some some experts worry that this could lead into some kind of addictive behaviors right You using an AI agent in ways that you know could replace some of your connections with you know that you talking to or texting with your friends

So I feel like adding this kind of adult content, you know, has some experts concerned when you're thinking about user engagement and addiction. But on other sense, I also talked to security and privacy experts who say that this could really make Muse accounts powder kegs for hackers who, if you're interacting in these kind of almost pornographic ways, that could be used to extort you, to embarrass you publicly. So whether it's privacy or just usage addiction, I feel like there's a lot of potential concerns to really dig in here and unpack better. I can't help but think of some of the stories that Wired has run over the past year about, you know, AI boyfriends and husbands and these full relationships. And again, this is like in some ways like a slightly insidious way to sneak that into like, don't worry, I'm just going to send out calendar requests for your meeting. And I'll say to you like they, you know, they say, look, this is for grownups at the same time, which you've also written about the mascots or the sort of animated interpretations of these agents

is as cuddly-wuddly as you can get. It is very much designed to be friendly and appealing to a broad range of ages, let's say. It looks like it is for kids, just so to say that. Youth advocates that I've spoken to for previous stories were concerned about this being appealing for kids even without the sexting element. So I do think there is, I feel like I keep saying the word contrast, But it's just not what you expect. You see this, like, LeBubu mixed with an Ewok, and you just don't expect it to get as freaky as it did. So I feel like this is, it really does not match with its kind of cute, cuddly mascot. Reese, thank you so much for joining us. This was great. Thank you for having me. Thank you so much, Reese. Everyone, please go read Reese's story. And if you want to see the sex that didn't make the story, I don't know, send us a DM. We'll see what we can do.

We are going to Nevada for this next story, specifically to a shopping center in the small city of Sparks. If you drive up to this strip mall, you're going to find typical shops, a Spirit Halloween, a hearing aid supplier, a Vietnamese restaurant. In this particular shopping center, though, you would also find the consulting agency, Lantern Production Consultants, LLC. This particular agency has been paid around $20 million since last year by Think Big, a pro-AI super PAC in the Leading the Future network. That's the network backed by AI industry figures like Marc Andreessen, Ben Horowitz, and OpenAI President Greg Brockman. It's pretty unusual for an agency to just exist like that in the Nevada shopping center. And the company also doesn't seem to exist anywhere else, though, beyond this post office box. Wired's Seth Owen found that the company is just one of four mysterious Nevada LLCs that have received payments from super PACs in the leading the future network. Election watchdogs argue that these LLCs are actually shell companies designed to mask the identities of strategists, consultants, and production companies working with the network for the midterm elections.

Brian, this was kind of a crazy story. Yeah. So, and I, you know, I am not, I'll admit it, I'm not steeped in election finance things. But anytime I see shadowy LLC, I pay attention. And I think it's interesting, too. I think leading the future, just to back it up a little bit, is they are very much like, let's give money to all of the pro AI candidates, right? Any pro data center, no AI regulation. But I guess, Leah, my question to you is, like, how common is it and what does it say about the sort of sophistication or ambition or shadiness of leading the future? It's a good question. And unfortunately, it's not super uncommon. At the end of the day, Tess figured this out by pouring through FEC filings and seeing that there were four packs that had the same address at this shopping center in Nevada. Like, incredibly shady, incredibly random. And in May, actually, the Campaign Legal Center filed a complaint with the FEC

alleging that Think Big and American Mission, as part of the Leading the Future Network, had broken federal law by failing to disclose the actual recipients of their payments. But this case isn't actually going to be heard anytime soon. The FEC is basically at a standstill. They have six commissioner seats. They're all supposed to be presidential appointees. And they need a quorum of four commissioners to make a decision on a legal complaint. They have not had a quorum in over a year and a half. So we're in this very weird situation where, to be honest, like there are rules. I don't want to say there aren't rules, but there kind of aren't ways to enforce them. So the fact that there are, you know, the fact that there is this network, this really big network, we're talking like over $100 million here of expenditures on big races, big campaigns. They have no oversight. And it's not overly cynical, Leah, perhaps, to say that the lack of oversight is a feature, not a bug.

I mean, it feels like the last thing that Donald Trump wants in an election is anyone paying too close attention to where the money's going. Absolutely. And when Tess reached out to Leading the Future, they sent an emailed response attributed to spokesperson Jesse Hunt, who wrote that the group complies with all applicable laws, and he called the complaint from the Campaign Legal Center merely a desperate ploy for attention from Public First, Anthropic, and the groups aligned with their outside organizations. Anthropic was not part of this suit. Neither was Public First. And so what this really said to us is that not only is this a way for leading the future and their candidates of choice and shell companies in Nevada to really kind of get their hands dirty with campaign finance. It's also a way for knife fights within the AI industry itself, for example, between OpenAI and big-time rival Anthropic, to be throwing punches at each other as well. I love it when random organizations and companies catch strays

when you're just reporting something totally separate. It's always a little thrill and always an indication that, to your point, yeah, there's a lot more going on here than we maybe even realized when we started reporting these out. We're only a few weeks away from the midterms now, Leah. Countdown begins. Countdown began. How big of a role are these sort of AI-oriented PACs playing in this election? How big of an impact are they going to have? I just remember, and listeners might too, like crypto PACs actually had a pretty decent impact on the previous election cycle. They had a ton of money and they spent it in smart places. Are we seeing an equivalent with AI or not quite or more so? Help me think about how to wrap my head around the degree to which they are impacting this election. It's a really good question. And I really want to make sure that we get across to our listeners here is that you have this network donating to races on either side of the aisle,

which is a little bit more unusual when you're looking at, for example, Fair Shake, the major crypto pack of elections past. It felt very, this is Republican, this is Democrat, there's no split in the middle. What these pro-AI super PACs are doing right now is something that's a little bit more interesting to me, which is future building. They are picking candidates that are perhaps not even in the most competitive seats right now for Republicans or Democrats, but they are picking candidates that will be able to become their biggest allies in Congress And so that has way further impact than just this election That has impact beyond 2028 We talking about like shoring up a support for AI in the United States government for honestly, without sounding incredibly dramatic, but like generations to come to be able to like make these viewpoints, whether that's deregulation or very light regulation or being pro data center,

if like that's how you're able to swing it in all these different ways. It's a lot of future building right now. All right, Leah, well, it's time to go from your favorite topic to your least favorite topic. We're going to do some consumer tech. Zoe is like somewhere in the bay, just absolutely thrilled to not be discussing this with us. I'm going to call her. I am going to call her later today and just make her talk about it with me. Okay, good. Yeah, that seems fair. This week, Apple announced some new products at a showcase in New York. Look, Leah, there's an Apple TV 4K. There's a new HomePod Mini. There's a lineup of LG built locks and cameras and doorbells and thermosets. I am not going to make you talk about any of those. Should I get any of them? I think you're fine. Okay, great. The one I am going to make you talk about, or at least make you listen to me rant about, just for a little tiny bit of time, is the Apple Home View.

So this is Apple's first smart home hub. By that, I mean, you know, Google and other and Amazon have had these before. So this is like their Alexa? Sort of, like with the display, right? So I have a Google version of this in my kitchen. I use it to set timers and to watch baseball games while I'm cooking and to play music, right? It's like a sort of, it can display a calendar, whatever. So Apple's is called the Apple Home View. It's a 7-inch display, sits on your counter, or you can get a version that mounts on your wall. It's not crazy expensive for an Apple product, which is fine, neither here nor there. You can recognize people by their faces, runs on Siri AI. All that's fine. Here's the thing. It looks like an Apple Watch. It's like a big Apple Watch type screen. So it's a square screen. And it doesn't play video. There's no video apps. You can't pull up, at least not yet. So it's shaped like a square. The aspect ratio is weird to begin with. Right. But then it doesn't have, like, Netflix or MLB TV, which is what I would be interested in, Prime Video, HBO 9, all the things.

Like, I feel like if you have a display, you probably want to watch things on it. Well, also just even, like, in this era of if we're talking about cooking, you're able to access something. This is supposed to be in your kitchen. This is supposed to be in home areas. I don't know. I would want to watch cooking videos on, like, Bon Appetit or TikTok. Like, what is the point of this? 100%. We asked Apple about this. I forced our gear editor, Julian, to ask Apple about this. And their answer is basically, look, you know, if developers want to make CarPlay apps, then they'll show up on this thing. So there's a world in which it could happen. Or you can always go through Safari and type in hbomax.com and log in like a dweeb. Like a dweeb? Like a dweeb, exactly. Absolutely not. So, look, I think this is an interesting device. I'm glad, like, Siri AI is better now because they're working with Google to make it better. People will be surprised at how useful it is. But the fact that it is like, yes, you can play Spotify, but you can't watch anything on the screen. That's what a screen is.

It's what it's for, Leah. I'm a little bit confused here because I have to say I'm sort of an Apple stan. Like, I love Apple products. I really, really do. And I have never interacted with an Apple product. I've said, oh, you know, I could have skipped this one. I think I could skip this one. I think I could too, which is I actually was looking forward to it. I kind of like, yeah, I don't love the Google thing I've got. Like, it's fine, but it's pretty old. It doesn't work with Apple Music. It's just like there's a lot that I – but I don't think I can commit to this unless, you know, developers get on board. But I don't know why they would. How much AI is this using? Like, this is obviously requiring it if it's a, hey, can Siri play me whatever music station? It's got Siri AI baked in. So basically, in that sense, it's like an Alexa. It's a little bit smarter, but it is. You will still probably, I think it will be just better at doing those things and better. I think there's a generation of devices like this where if you asked it any question other than what time is it, they'd kind of freak out and show you a web page with mixed results.

This should be more conversational. It should be able to actually give fuller answers to things. But I don't know that it's, I don't know that you would want it just for that. It's not like it can do something different than your phone can do, and your phone's already probably nearby. Yeah, no, I would just prop my phone up. Sorry to hear that. Sorry for you, Brian. Thank you. Thank you for just letting me, you know, have a safe space to let my feelings out. Twice a year. All right, Leah, before we go to break, I have a quick quiz for you. I'm going to read four names from the AI industry. And Leah, I want you to let me know how many seem familiar. Are you ready? Yep. Okay. Sam Altman. Familiar. Mira Mirati. Also known who Mira Mirati is. Susan Milner. Nope. Craig Dalton. Nope. Wow. Well, Leah, it makes sense that you only recognize Altman and Mirati because Milner and Dalton don't exist. You know who insists that they do exist when asked the same set of questions?

I'm going to infer, based on the title of senior writer Kate Taylor's excellent story, men. Men, Brian. Men. Men. It's true. It's men. In a new survey of more than 5,400 Americans, Harvard researchers found that more than half of men claimed to know these fictitious AI luminaries. Women also did, but to a much lesser degree, so around 40% of women. So by a sizable chunk more, men are super confident with all the wrong answers. Does this come as a surprise to you, Leah? I mean, of course not. And there's a lot more to expound upon there that is probably not like incredibly scintillating. What I would say, though, is I'm fascinated by like, you know, a confirmation of biases here, But also fascinated by the fact that like this is this is such a new industry that, of course, there is a not insignificant group of people saying, I already understand every single thing about it.

Like, I've never been more included. No one knows more about the 1993 player of the year MVP than I do. Like, that is such an easy jump. Tell me tell me how this was all found out. So Harvard University study, and I think it's important to say we led with this tidbit because it is so fun and damning, but there's a lot more here to it. It was actually a pretty extensive survey. So Laura Manley of the Harvard Kennedy School Shorentine Center was involved in the study, told Wired that they're seeing that men are more optimistic and more confident across the board when it comes to AI. That, of course, veered into overconfidence in this case. We found that across a year of media coverage, men were named in headlines about AI about 14 times as often as women. And part of that is a function, too, of obviously the top AI executives skew heavily male. But it also includes AI experts, men to be quoted as AI experts a lot more than women. I think the whole point of the study is, you know, where are the sort of gender dividing lines in AI?

How is this industry as it continues to transform the world? it does appear to be transformed the world differently depending on if you're a man or a woman and your attitude towards it kind of reflect that. I wonder if so much of that has to do with not being concerned about the guardrails, right? I'm going to have to delve into this study myself. I'm going to make Kate tell me all about it over lunch or something. But there is something very fascinating to me about when I'm thinking of some of the biggest critics of AI. I'm thinking of like Timnit Gebru who we had all of this excellent coverage on in the last few weeks I thinking of people who are not just like tearing the industry down but saying like hey here are the issues here And you obviously like have some of that from the very top We been talking about Anthropic We been talking about Amadei CEO of Anthropic But there does very much feel like that gap to me. It's really interesting. I'm fascinated by it. No, I am too. And I think it is something that the industry does need to address and be aware of.

I think there are it's also all ties into, I think, with, you know, Wired senior correspondent Lauren Good wrote a feature last month for our Wired Women issue about who's the most powerful woman in tech. And there's not really an answer because there aren't these sort of cults of personality that are built up around, even though there are several very powerful women in tech and in AI, they don't have this sort of, you know, gravitational pull for better or worse, often for worse. that the Sam Altman's, the Elon Musk of the world do. Coming up after the break, we are sharing our Wired Tired Picks of the Week. Stay with us. It is time for our Wired and Tired segment. Whatever is new and cool is wired, and whatever passe thing we're over is tired. Brian, what do you have?

I'm going to go on faith because I have not seen this movie. But people I respect have seen this movie, and they love it. Is it Digger? It is Digger. I'm seeing it tonight. I'm so excited. I'm curious what you have to say, but I'm going to say that Digger is wired, if only because, I think even not having seen the movie, I'm going to say it's wired because it has sparked, I think, a really interesting conversation, not just about Digger, but about how we talk about climate. Whether you like the movie or not, whether you thought it was effective at communicating that, as senior writer Molly Taft does at Wired, or whether you think it's garbage, as many other people do, it has at least forced a conversation about how we talk about climate. And I think we're in this place where so many people seem to have moved on from climate conversations. You've got the sort of AI extremists who say, oh, that's a solved problem. It's fine. You've got the sort of MAGA contingent who says, oh, it was all overblown. Why are we worried about it?

I think climate has left the conversation from a couple of angles. So whatever you think about Digger, the fact that it is it has got the conversation move beyond the movie itself into how we talk about climate. I'm supportive of that. That's wired for me. And I'll use tired. I'll use my tired to complain about the Apple Home View again. I think you got to you got to figure that out. You hate it? I don't hate it. I just, I'm not mad. I'm disappointed, Leah. This is a plea to anyone that works at Apple. Hear Brian's prayers. Help him out. Okay, those are pretty good. Mine are so internet-y. It hurts. I feel like you're going to know exactly what they are and could probably say them for me. My wired is that Girls is coming back. That was my guess. I'm so excited. Oh, my gosh. It is Millennial Catnip. Hannah Horvath and Marnie Michaels will be revisiting our screens at some time soon. HBO, this girl's revival is everything I care about.

I really, really can't wait, as you know, because I messaged you about it so many times yesterday. Which one are you? Well, not playing that game. Absolutely not. This is not the sex in the city. Who's Carrie? Who's Samantha? all of the characters of girls are equally amazing and terrible. The only thing that I will absolutely say is, like, obviously you want to be Shoshanna and you do not want to be Jessa if you've rewatched it in your 30s. That's good advice. That's actionable advice. My tired is also incredibly Internet coded, which is I let me let me take you into the corner of my Internet and my algorithm, which is the Kelly Clarkson bird debacle at the Magic Store in Brooklyn. That's everybody's corner. That's everybody's corner of the Internet. I actually don't think it is because I tried explaining this to my husband yesterday, and he was like, are you having a stroke? Oh, no. What are you saying? You've got to get on board, Rodrigo.

Come on, man. You've got to get on board. Basically, real quick for our listeners, Kelly Clarkson, noted singer and talk show host, walked into a magic store in Brooklyn and tried to buy an animatronic bird named Molly for $15,000 and was told, absolutely not. Molly's not for sale. The back and forth has been incredible, just truly incredible. Kelly Clarkson posted about them and the store returned. According to an interview in New York Magazine yesterday, the store is getting harassed. Kelly Clarkson has doubled down. There's a bird animatronic bird maker that lives in the woods upstate who only has made two of these. It's a core component of the magic show. Anyway, it was like really fun for me for I think about like an hour or two and now I'm over it. So that's my tired. I think much like girls, much like girls, you don't want to be anyone in this situation. I don't think anyone

comes out looking good. I think it's all a mess. Except for Molly, the animatronic bird. Except for which does look pretty cool. That's our show for today. We'll link to all the stories we spoke about in the show notes. Adriana Tapia produced this episode. It was mixed by Cran Bandy, who's also our New York studio engineer. It was fact-checked by Matt Giles and Daniel Roman. Mark Leda is our San Francisco studio engineer. Kate Osborne is our executive producer, and Katie Drummond is where it's global editorial director. Thank you.

execute, access, and connect to. Set clear boundaries around the applications, files, systems, and online destinations your AI tools are permitted to use. Because a proven AI tool should not mean giving it unrestricted access to your environment. Control what runs, control what it can do, control what it connects to. Learn how ThreatLocker can help you safely embrace AI at threatlocker.com slash wired. That's T-H-R-E-A-T-L-O-C-K-E-R dot com slash wired. 150 years ago, they were hunting us down to kill us, and now they're hunting down immigrants to deport them. This is First America, the true story of how the United States came to be and how we got to this present moment. Listen to First America wherever you get your podcasts. Hi, my name is Zach Barron. I'm the Senior Special Projects Editor at GQ,

and I am pleased to introduce One More Question, GQ's new weekly interview show. Over the years in the magazine, I've written dozens of cover stories and traveled the world to talk to and write about some of the most interesting people in the GQ universe, from Martin Scorsese to George Clooney and Brad Pitt to Outcast Andre 3000. And now I'll be doing it weekly. I'm going to sit down with some of the most interesting people in the world and figure out what makes them, well, them. New episodes of One More Question will be available each week wherever you get your podcasts or on YouTube if seeing people in chairs is more your thing.