Ben Whitelaw 0:00

Jeff, just because you're a co-host of Ctrl-Alt-Speech and a guest of ours, doesn't mean you don't get subjected to the same question that I ask Mike every single week, which is, what fun thing have you seen on the internet this week?

Jeff Horwitz 0:10

All right. Two answers here. the most basic one, is just that there is, uh, somebody on Twitter, I guess X, that, is posting regularly mazes they've built for puppies, where puppies try and then hilariously fail to get through extremely basic mazes to a pile of treats. Uh, like they are up to the 28th configuration of the mazes. it's beautiful. It's beautiful. but the thing that I would cite that I, I really loved, was, again, this was on, X. a woman basically gave her dot, agent, the ability to go plan a evening with a friend, and, just basically call all the shots, like full control, wasn't gonna veto. She was just gonna like do what the bot said. And, this went hilariously awry. I mean, like it had some very good, very good impulses. it also like badgered the hell out of a restaurant for absolutely no reason, and then decided that it needed to make the menu gluten-free, despite the fact that that was not a thing that like was even remotely suggested. And I think it's like, while hilarious, it's also an indication of where we are currently going, which is like we're just gonna trample over a whole bunch of social norms. Namely, like when someone emails me historically, unless it's like a spam marketing email, I assume that someone like took the time to interact with me as a human being, and that is like not a safe assumption going forward. And, and I don't like that world. I really don't wanna live in that world, but that is definitely the world we are all going to be living in, thanks to the wonders of technology.

Ben Whitelaw 1:44

Yeah, it, indeed. What a, what a, broad gamut of, fun things you've found this week. I wasn't expecting puppies in mazes, to be honest. and so that's, that's a very nice image you've conjured for us at the start of Ctrl-Alt-Speech. my kind of fun thing, just for your, benefit is, actually kind of much more mundane than any of those. It's a, a newsletter that I've discovered called WikiPop, and it is a, a really kind of quite perfect daily email that you get sent of the most visited article on Wikipedia the previous day. And it's kind of like, it's perfect in so many ways because it's, it's social proof at a Wikipedia scale, right? It's, the best things about the internet for me. It's like community-driven. It's Wikipedia, which I love, and, you know, should be protected at all costs, and it's informing in a way that, many of the feeds that I inhabit otherwise are not. So, I would recommend WikiPop if, if you don't already subscribe.

Jeff Horwitz 2:33

Oh, I, I like that. That's nice. And yeah, no, Wikipedia is, uh, one of the, internet nonprofits that I, regularly give money to out of guilt. They just like get me every, every three to six months with the like, Hey, are you sure you're paying your share?" Uh, you know, note. And I'm like, "Ah, hell." Like, it's been a while.

Ben Whitelaw 2:52

It's got me. It's got me again. Hello, and welcome to Ctrl Alt Speech, the podcast where we make sense of the major debates shaping online speech, platform power, content moderation, and the future of the internet. it's October 22nd, 2026, and this week we're talking about the US state going after Meta's bad ads, the promise of AI audits, and for our extended Patreon episode, TikTok's child safety reckoning. Ctrl Alt Speech is the never heavily advertised podcast of Mike Masnick, founder and editor of Techdirt, and me, Ben Whitelaw, the founder and editor of Everything in Moderation. But this week we're joined by a very special guest, Jeff Horwitz. Welcome to Ctrl Alt Speech as our guest host for this week.

Jeff Horwitz 3:39

Oh, thanks for having me

Ben Whitelaw 3:40

for those who don't know Jeff's work, which many of you will do, Jeff is a, an investigative reporter and author who's currently working for Thomson Reuters, but whose work at Wall Street Journal is very, very well known. He was the leading reporter on "The Facebook Files," which was a tranche of reporting that was based upon the leaked documents of Frances Haugen. And if you've heard me talking on the podcast or writing on, on Everything in Moderation in recent weeks, that work was central to the recent film, "The Social Reckoning," which, features Jeff being played by Jeremy Allen White. Which I know you've got a lot of questions about previously, Jeff. Uh, I'm not gonna ask you those 'cause they're, very much kind of below the level of this podcast.

Jeff Horwitz 4:25

He's got better abs

Ben Whitelaw 4:29

I wasn't gonna ask. Um, and we're not gonna, try and prove that. I, I do wanna ask though, like, have you been reading the reviews of "The Social Reckoning?" and what has the last couple of weeks been like as the film has come out and, people have been talking about it?

Jeff Horwitz 4:40

Reading reviews, no. I tried to keep myself sort of distant from this process. I was extremely excited it happened and really flattered that this was something that, like, anyone wanted to try to make a movie about. and, I had some early consulting just in terms of like my experience reporting it out early on. But the movie was fictionalized and, while I did attend the premiere, I've sort of like tried to keep movie Jeff separate from real Jeff and, part of that has been, trying not to indulge too heavily in, in reviews. my general read is that they have been, um, mixed, so

Ben Whitelaw 5:18

I think that's, I think that's fair, but I think that, you know, to be talked about at all in this day and age is, is a good thing. And, you know, I, I wrote for Everything in Moderation that this is almost like media literacy for millions and millions of people.

Jeff Horwitz 5:28

it is, and I, I think that was like one of the things that was really exciting about, the interest in this from Sorkin was that like, look, I spent seven years writing about Meta and, the choices that it was actively making that governed the experience of billions of people and what was talked about, what was shared, what perceptions of reality were. you know, so I was blue in the face. And so I think it, was like really enticing to have someone sort of take a fictionalized approach to this to present it to a different audience. And like, to some degree, like If what people who are not exposed to this world day in and day out take away is that actually platforms do have massive control, over what happens, not necessarily by like, you know, moderating each and every post, but simply by how they have designed their systems, that I feel like would be a big win. because like, I just can't, go on if we're just gonna keep on talking about well, you know, how's this different than, AT&T? They carry phone calls. They're an information provider. Like I just, I can't, I can't do that. So I, I am, delighted that, there is a movie out there that sort of makes that point even if some of the details are, fictionalized.

Ben Whitelaw 6:40

Yeah. But it makes conversations with family and friends a bit easier now that that film is out in the world.

Jeff Horwitz 6:45

Little bit, little bit, yeah

Ben Whitelaw 6:46

I wanted to ask, obviously kind of still on the technology beat. You're working at Thomson Reuters now. We're gonna talk a bit about some of your reporting over the last year, still on, on Meta, but I think with a slightly different focus or slightly kind of narrower, focus maybe. how do you kind of think about covering Meta now after seven years, you still as kind of resolutely focused on them as you were? How do you think about the kind of social media era now that we have these AI behemoths emerging as well?

Jeff Horwitz 7:12

Yeah, I mean, I am trying to, uh, broaden significantly beyond Meta these days. it's been a long time, and, uh, I have been, um, kind of a one, one-trick pony in terms of the corporate reporting. mean, mind you, it's been a great trick, and I, I just, I just love, I, I love the trick. Um, but, Look, I think there's a lot more, similar than different, in the new age, right? you know, Meta was formed by basically, a land grab in social media, and in particular, a land grab you know, with the new form factor of, phone-based content. and, LLMs and the devices in which they operate are providing the exact same kind of Cambrian explosion of, new services and technologies and features. And, they're being built in the same extremely rapid pace, where, it's kind of a, you know, winner-take-all, first-mover advantage, set of pressures. And people are doing things that I think, as they are saying themselves, as they are doing them, they're not wholly comfortable with. so I mean, to some degree, maybe we're, like, a little further ahead than we were with the social media era in the sense that, like, Meta circa, 2014, 2015, it was like, "We just make the world be-better by connecting people all the time." Whereas, the current, crop of sort of new actors are sort of along the lines of like, "We think we might be making the world better, but, we could be dooming us all," right? And, like, that's not the most, I think, useful frame. To some degree, I'd, I'd like to see the AI industry get a little more granular and less ex-risk focused. But, like, at least there's, like, an openness to the downside that isn't going to require, you know... people are going to recognize bad things when they happen, I think sooner than they did in the social media era.

Ben Whitelaw 8:53

Yeah. I, I've been surprised by how quickly we've moved into talking about AI safety, so soon after the kind of technology has emerged. And like you say, there's, different tenors to the conversation, but, it's out there and it's, it's kind of front page news. I wanted to delve into one of the things you said on, on X/Twitter, a few weeks ago about how AI employees are talking about these safety concerns in a way that was similar to folks at Meta. and you kind of almost feel a sense of deja vu about that. what are the implications of that, do you think? And what, what might we learn from your experience of tracking the, concerns of Frances Haugen and other whistleblowers?

Jeff Horwitz 9:30

Yeah. look, I think that, the way that I have traditionally gotten sources, inside Meta was people who are working on things with giant societal ramifications. In many instances, like, their job is to guard against certain bad outcomes, but they're not the center of power in the company. they get overruled. And there is, no faster way to radicalize someone than to tell them that their job is to prevent a calamity or harm to human beings. pay them really well for that, incidentally, and then disregard when they tell you to stop or to change course. and the difference I see right now is that, like... And I think early on, like if you go back to like 2016, 2017, there were definitely plenty of people saying, What Facebook is doing is not great. I have great qualms." It was in the same way that people leaving Anthropic and OpenAI keep on saying things like, I have serious reservations about how management is proceeding." and those things back then, as now, did get a lot of attention, but they're kinda dead ends. I think the thing that, I realized with Meta reporting was particularly in a kind of degraded news ecosystem. To really get past the, like, kind of he said, she said element of things and to hold attention, you needed granular examples of the specific trade-offs, and also you needed to demonstrate that the company understood what was happening. It can't just be that bad things happened, whoops. it has to be okay, like this is the cost of doing business, element of things. 'Cause if you learn about thing and you're fixing it, and you're fixing it, I mean, that was for years, that was Meta's line, is like, You know, we always try to do better, right? And I think that one of the things with the Facebook Files was like, some of that stuff undercut the idea that they were always trying to do better. Uh, it actually turned out that they were doing as well as they wanted to, thank you very much, and actually growth was more important. So I think kind of the definition of what a whistleblower looks like is, in the AI field, maybe still evolving. and a part of me is a little bit like, wake me up when, people are, willing to discuss the specific, shortcuts taken in post-training, that, they consider problematic. I mean, I think like something that like And you could tell that OpenAI got very concerned when there was the, leak that, basically that they were reducing monitorability, to boost performance, uh, is the short version of it. Like that particular thing, That they care about. People saying like, "I am gravely concerned that the technology might end the world," that's not something that gets you fired, it turns out. Uh, it's like, giving away the detail that they have gone back on something that in previous blog posts they've said is really important. that's something that, know, I think the companies kind of rightly are more sensitive to.

Ben Whitelaw 12:26

Yeah. Interesting. One of the stories we'll talk about today actually is, is in that discrepancy between what companies said publicly and what they did in practice, which I think is why it's such a great week for you to be on the podcast talking about these things. we could go on and on about, your likeness to Jeremy Allen White or, you know, all of those other things that I'm sure listeners wanna talk about, but we're here to talk stories. so I wanna, get us into those big stories this week, we're very lucky actually that just before we started recording, your latest big story for Thomson Reuters it's been published about Meta, and it's, that kind of longstanding story about scam ads that you've reported on. If people don't remember, there was a really great piece of reporting that many people who I speak to refer to, about how Meta's 2024 revenue was essentially gonna be made up to, a significant degree by revenue from scam ads and ads that were based upon banned goods. And so you, found this out, Jeff, from previously undisclosed documents. Before we get into this new story and, and, what a state in the US is doing about it, do you wanna just kind of explain why you decided to focus on ads as a particular point? Because it's not something that your average, Facebook user or, news reader will necessarily care about, but it's how the company makes

Jeff Horwitz 13:41

to engage in a little bit of self-criticism here, the fact that it took me five years of writing about Meta, to really get into the core of the ads business and the choices being made there is, um... I can't say that's flattering to me. Um, I, did a, a, lengthy project last year, we won a Pulitzer, yay, on, how Meta approached, scam advertising. and the answer was not by, harshly cracking down on it. turns out that they were, doing, about $7 billion a year in what they, refer to as higher legal risk scam ads. in other words, stuff that they, like, could credibly be detecting and it looked bad. They were hiding that from regulators, or at least, I'm sorry, they were, they were making it harder to find in the ad library, was a stated goal. they, were leaving up advertisers unless their systems were more than 95% certain that, the advertisers were, fraudulent. so like 94% was not an issue. They were charging a premium to discourage scammers. So in other words, they'd actually charge presumed but not certain scammers more money in ad auctions. and, they were also allowing, their business partners in China to, like, sell whitelisted, so enforcement-protected accounts to basically anyone. we actually demonstrated that by, buying some, protected whitelisted accounts from these Chinese intermediaries and, then running, extremely scammy ads with, great success. so it was, kind of an indication that where the company was at in terms of scam enforcement, particularly at a time when, like, our knowledge of scam enforcement and how gruesome it is in terms of overseas scam compounds, you know, if anything, getting your savings ripped off in the US is the much better version of that. So, you know, you're on the luckier side compared to the guy that is, literally getting cattle prodded if he doesn't manage to rip off your savings, in a Cambodian scam compound. So, they were not doing perhaps all that many outsiders might have assumed they should do. and that was kind of the, gist of the reporting.

Ben Whitelaw 15:47

And it was, it was very well received, it's worth saying. Not, not just because it was a Pulitzer winning, but, a lot of trust and safety professionals who I've spoken to, read it and, got in touch. there was a big conversation at Trust Con, the big conference of trust and safety professionals, in the summer. So it was really a great piece of reporting. And, and today, Jeff, uh, you were kind of rushing to get this tied up before we started recording. But today sees a, the first US state you believe to take that reporting and, bring a case against Meta for it.

Jeff Horwitz 16:15

Yeah, Rhode Island just sued. so, it does draw significantly on the documents that, Reuters obtained. and, you know, that's always gratifying to see. Uh, that said, there's definitely some angles in there that, clearly they are, moving in discovery already past things that I could obtained. There is a suggestion that perhaps, some of this stuff is tied into a tolerance for fake accounts. there is also, some heavily redacted material, related to, how the banking industry, responded to this stuff. And, it seems like, there is, I, I'm looking forward to seeing that stuff, if it ever gets unsealed, a great deal as, you know, I heard, rumblings that, Meta was, shall we say, extremely aggressive in pushing back, on financial institutions that were, angry that Meta wasn't doing enough to protect their customers.

Ben Whitelaw 17:06

Interesting. And so, so this is the first state to have gone down this route. I mean, Meta has, not commented on this particular story. It's very, very fresh. are you expecting there to be other states that take a similar approach depending on, the kind of outcome of this?

Jeff Horwitz 17:21

I don't want to forecast these things. I will say that when I was doing this work, on scam advertising, I was kind of comparing it in the back of my head to previous work I did on child safety that was fairly... Some of it was fairly central to the state AGs' actions, against Meta. Like, they opened up the trial, the class action or the, uh, the mass state action, that recently settled, opened up with Arturo Bejar, who was, uh, a whistleblower I, I took public. and I kept thinking to myself, if like this is something that is gonna lead to years of litigation and billions of dollars on the child safety front, it doesn't even involve money. This stuff is like very directly involves consumer harm and in some ways seems like it's almost maybe even an easier lift to make an argument about in terms of that, you know, Meta has done a thing that benefited itself at the cost to users. So I, you know, I don't, I don't wanna like predict it. I will say that Santa Clara County has also, filed on this, independently. they were the, the first to do that. and you know, I, I wouldn't be shocked, again, given sort of the timeline for sort of, mass litigation, I wouldn't be shocked if there are other states that come out of the woodwork. I don't have a specific knowledge of anybody that's about to file or anything.

Ben Whitelaw 18:41

Got it. I mean, it's, it's a really interesting time for Meta, not least because we'll come onto a story, about the, another case that many people will know about in a sh- second. But they have seemingly done a lot of work over the last year and a bit since this story that you published came out to combat scams and fraudulent ads. you know, you don't have to go that far back in its kind of newsroom, press release system to see, its kind of public, positions on the work it's done. there was one post from earlier this year that said, "We've removed 159 million scam ads."

Jeff Horwitz 19:15

Oh, nominal, large nominal numbers. The friends of tech platform t-, uh, press releases. They love them. Yes

Ben Whitelaw 19:23

Exactly. And it's, it makes a g- for a good headline. You know, there's, been a clear concerted effort to show themselves to be kind of on top of this. do anybody who's kind of used the platform either as an advertiser or as a user can probably see that it's not improved a great deal. What are your senses of, how they've gone about trying to improve the ads space on the platform?

Jeff Horwitz 19:43

I, I-- look, I can see the announcements made. they have, certainly, had some very public crackdowns on, specific things. I will say that, it does appear from what Rhode Island is alleging that Rhode Island is having absolutely no problem finding many of the exact same problems. the thing that I think many Meta people who are working on this internally flagged as being, a foundational first step, to combat scams and something which I might add, Google has basically largely done for pretty much all advertisers, which is verification. it's really hard to police scams if you don't know who you're taking money from in the first place. And like, this is something that say, in the financial services industry, you don't sell financial products to an entity if you don't know who they are. that's just like, it's just not done. ad sales, on the other hand, there are no such rules. And so like, a problem that Meta was having repeatedly per the internal stuff was, you know, they'd, boot bad advertisers and they'd just go right back to the same Chinese intermediaries, buy some fresh accounts, and be right back on the platform that same afternoon. And so I think what I would wanna see, and if I were a regulator, I'd wanna see would be what Meta is actually doing, not in terms of like nominal takedowns, but, okay, what are the thresholds for, removing an account? how many, predicted scam ads, is Meta, allowing, per country? cause they, I mean, they absolutely have classifiers that, do that. you know, what are they doing in terms of using the signals coming out of scam compounds, for enforcement? Because there's absolutely, signatures of that sort of activity. There's like a lot of sort of granular stuff that is much more process-based than we took down a whole bunch of accounts or, you know, we've committed to a policy or, things like that.

Ben Whitelaw 21:36

Right. Interesting. So the discovery, is gonna be very telling in that regard. I mean, it, it comes in the same week, Jeff, as, many people will know, just yesterday in fact, that New Mexico I mean, a New Mexico judge has ordered Meta to pay $25 billion for lying about, the, use of data in the, the wake of the Cambridge Analytica scandal and, misleading the public about what the company did as far as handling misinformation and hate speech on the platform. New Mexico is the one state that didn't settle in the big settlement that you alluded to, a second ago. And so a jury found Meta, guilty in, September, and the judge has since been working out how, it would pay, and we've landed on this 25 billion number. it's funny just to kinda think about how they arrived at that.

Jeff Horwitz 22:27

Well, the, the judge, the judge said that it was like, should have been well over $200 billion if they were strictly applying the law. But like, I guess he was gonna cut them a break. Um, but I don't think Meta sees it that way.

Ben Whitelaw 22:38

No, no. I It's still one of the largest fines leveraged against a U-US corporation, as far as I understand, for something like this. And it's, you know, it's more than the 18 billion pounds that the, multi-state settlement concluded.

Jeff Horwitz 22:51

You said pounds, not, uh, dollars.

Ben Whitelaw 22:55

my Britishness getting in the way there. Um, that would be even worse, I

Jeff Horwitz 22:59

Yeah, exactly. Yeah

Ben Whitelaw 23:01

but it's a big moment in the sense of, you know, again, further kind of entrenching the idea that Meta, were misleading the public in a number of different ways as to what they were doing or prepared to do, which I think, I guess, is kind of what we have seen, in other cases as well

Jeff Horwitz 23:15

Yeah, I think, one takeaway from this, and this applies to the child safety stuff, it also applies to the scam litigation. is that the most dangerous words a platform can possibly say is, "We are handling this responsibly." if you say that publicly and it turns out you are not, then you have opened yourself up to what I think is increasingly being viewed as a potential end run around 230 protections. which is that, you know, your platform can be as awful and as, you know... how awful your platform is doesn't matter, but if you're saying that it's safe, you are misleading consumers and therefore you are at risk. So like, I mean, honestly, like the legally safest response to any criticism, from a platform does appear to be it is what it is. I don't know we're gonna seeing that in, press releases anytime soon, but like, you know, Cambridge Analytica obviously, uh, it never dies. our children will be discussing, ongoing litigation involving Cambridge Analytica, I suspect. but, you know, I think, yeah, it, it's, uh, a lot of this is always comes down to, what the company said, its standards were versus, what, a court might later find them to have been at the time.

Ben Whitelaw 24:21

Yeah. And, and this kind of growing trend to see platforms as, products, as, products with risks and, litigating them- against them a- accordingly. there's probably not too much more to say on the No Me- New Mexico piece. I think it will, will have,

Jeff Horwitz 24:35

I mean, that thing's gonna get challenged. I, When I saw it, I was like, wow. One, you know, New Mexico really seems very comfortable going it alone and seems to be, like, pulling down results that, if I were another state's AG, I might be a little jealous of. but like, it's also... I, I don't know. I, I sometimes have a hard time imagining that, $25 billion, is going to get paid to New Mexico's schoolchildren, which would be a, uh, You know, those are gonna be some really nice schools given New Mexico's population, if that is in fact the case.

Ben Whitelaw 25:04

Yeah. Yeah, their budget is, I think, a s- a number of billion dollars below that

Jeff Horwitz 25:08

Yeah, the, the, and the, the entire state's budget is, yeah, is far less.

Ben Whitelaw 25:12

Yeah. so, so, maybe it's time to move to New Mexico. Um, is worth noting the kind of New Mexico Attorney General, Raul Torres, he's on a bit of a run. Um, I think in, in the States you call it three to zero, three to O, in sporting terms. But, you know, there's been two other cases this year that New Mexico have won.

Jeff Horwitz 25:29

And they've got a lot outstanding still against TikTok, Snap, they are by no means done

Ben Whitelaw 25:34

Yeah, no, it's true. so I, I wonder what other states will, will end up learning from, from their success. And, and the, the other thing to note is just how this has been kind of this judgment has been kind of created, made to be by the company, less of a consumer protection judgment, but a, a speech story. And I was noting about in one of the articles I read how Meta's chief legal officer called the result an attempt to impose a tax on speech to enforce new rules for what people can say online, which is a, it's a kind of felt like a bit of a dog whistle to the US administration, at the very worst, and at the very best felt like a kind of conflation of or an intentional conflation of what this judgment really is about, which is clearly about, data privacy.

Jeff Horwitz 26:18

yeah, I feel like we, we-- there is probably an angle in reducing any criticism of platforms, to, help their attacking speech. cause yeah, I mean, the, details of, Cambridge Analytica, and again, I, I sometimes get a little lost in how the damages occurred from the Cambridge Analytica leak, and like what the concrete harms were. but, that said, it's not a speech question. I mean, unless we're calling like shipping off user data in violation of your own terms of service as speech, which I don't think Meta is doing. But, you know, at a high level, yeah, sure, call it an attack on speech

Ben Whitelaw 26:54

Yeah. and what's notable about that, which is the kind of third story in this little Meta triumvirate, is the fact that that legal officer, CJ Mahoney, is gonna be playing a larger role in a world without Joel Kaplan, Meta's, chief public policy officer, who this week stepped down after 15 years in various public policy roles at the company. Kaplan is a, kind of Republican darling, I would say. W- you know, a, friend of the administration, a long-term, pal of Zuckerberg, and was given a kind of very warm send-off by the CEO of Meta in some of the kind of public comments. you can kind of read too much into some of these departures, Jeff, I think. But I think after obviously the revelations in Sarah Wynn Williams' book, which, were not proven or Meta disputed and, and Joel Kaplan disputed and also these New Mexico judgments, I wonder if there's a, an element of, somebody taking up the reins who just isn't Joel Kaplan.

Jeff Horwitz 27:52

I mean, I, I don't, it doesn't feel like they are, short of people who are friends, with the administration. There's, uh, Dina, Powell McCormick. There is, Kevin Martin, who is, the long-term deputy of Joel, is, you know, still very central to things. So, like, I don't, I don't know that I see this as, like, a, change in direction. Kaplan had been there for a long time, and certainly oversaw concerns that the company might be, needed to sort of get along better with the administration, with the current administration, during, both of President Trump's terms. I don't know necessarily that, like, he was always beloved by people on the right. I think, you know, there are people who are more vociferously outspoken. but, I don't know I'd, I'd read much into it either way, in terms of it feels like it is, uh, more of a internal intrigue than a shift in direction for the company.

Ben Whitelaw 28:43

Yeah. Fair. So in many senses, you know, those three stories I think demonstrate that, all is normal at Facebook. Everything is the same. Nothing has changed. We we, we, we continue as were, and you will still have, interesting stories to dig into at the big blue, I would say. we're gonna move on now, Jeff, to talk a bit about, a couple other interesting stories, some eye-catching stories that, caught our eye, in our feeds this week. and then we're gonna move on to our Patreon section for supporters of the podcast. let's start, Jeff, with a story that you saw by NBC News about YouTube in schools, 'cause there's, something really nice about the treatment of this story and also the, topic of it.

Jeff Horwitz 29:23

Yeah, I, I can't say enough positive about this story. and in fact, I'm, annoyed at myself that I didn't think of it, 'cause I've, I've done, I've done stuff like this before, right? Like, I think in a world in which, external access to platforms is extremely limited, you know, you just gotta throw sort of first-person testing at it. I've done this with Meta's sexy chatbots for kids. I've done this for, their recommendation systems, if you follow, say, gymnasts. uh, that was extremely ugly. Like, I've done this for sort of ad targeting things, for scams. Like, just actually using the damn platform is like shockingly not done by reporters or academics as much as you would expect. So like what NBC did here was just, they went to parents and teens, and they got, their YouTube histories, and then they matched up what they were watching, during school hours. and um, it is a-about as ugly as you would expect, perhaps. but Ben, you are like, what you would expect and what listeners would expect is perhaps different than what other parents might expect because, you know, Google has pitched YouTube as being an educational tool that really belongs in schools. And like it turns out that if you're looking at what is actually consumed, on, a small number of individual students', YouTube accounts during school hours, you're gonna see a lot more like watermelons being blown up with handguns and, lascivious dancing than you are going to see like explanations of, physics and, algebra. So, it just is like extremely basic and I just when reading it, I was like, "Damn, that's good," and also, how the hell is it 2026 and this hasn't been done, in like a clinical academic setting, previously? and like I, I just feel like it, that we're still kind of approaching platform stuff sometimes without actually like using the damn things like a normal person would or looking at how a normal person does use that data.

Ben Whitelaw 31:23

totally agree. And, you know, some of the best, some of the best reporting on, on platforms generally I think almost straddle that line into academia and it would be something that, a researcher would do, which I think is, is very telling. I mean, the, just the numbers, of some of these pupils is, is shocking. You know, one pupil, just to give an example, looked at 4,700 videos during school time over a year period, and, like, less than 5% were educational. You know, the numbers were, were shocking and, and the, the piece does a really good, job I think of visualizing that through some nice infographics and kind of talks you through it, again, in a way that I think is, is very powerful and that is likely to get noticed,

Jeff Horwitz 32:00

the same thing could be done, right? We are, like, still in the, we're in the middle of a national debate over, phone access in schools, um, at school. Like, There is no reason why one could not forensically audit what was being consumed at what particular times, in a cohort of children's devices, and try to address the question of well, is or isn't it a good idea for kids to have access to phone in schools? 'Cause I think we can, like, kind of generally all agree that you are probably not learning at your best if, like, Reels is currently streaming or, like, you are, actively taking Snapchat, you know, sending pictures on Snap or, you know. Like, this feels like it's, like, kind of a, maybe the result's obvious, but, like, if we need to have that conversation to discuss, the question of okay, are phones good or bad for education? I mean, there's two ways you can go about it, right? One, you can, like, run long-range causal studies, in which you withhold things. The other one is you could just, like, look at what the hell people are doing on their phones, and that is gonna give you, I would argue, a pretty good sense of, whether this is helping out or harming education

Ben Whitelaw 33:07

Yeah. and I think this will, this will be taken up as a story by parents and school boards and educators in, droves. I think this will inspire a really strong conversation. So

Jeff Horwitz 33:19

Yeah, it's just like, I just wish that like, somebody needs to do this work in a little more thorough of a fashion. And again, that's, that's some self-criticism on, on my part. I could have proposed this to academics and tried to partner with them on it. I've done plenty of those things, but I haven't done this. And like, yeah, man, field's wide open in, you know, year of our Lord 2026. what the hell are kids doing on their devices during school hours? It remains a, like, interesting unanswered question.

Ben Whitelaw 33:44

Yeah, exactly. reporting the better. let's move on to another, another eye-catching story now. It's, it's much more of a tragic case that, is worth talking about. we're gonna briefly mention suicide here for any listeners who, need to be told in advance. But there is a, uh, a story that's been, published, by actually a, an, a union for, online workers, the UNI Global Union, about the suicide of a woman in France who had been previously working as a content moderator for Meta in Barcelona, where there is a large outsourced moderation community and several large BPOs work there. this is a tragic story in which the family the, the moderator believes that her death was linked to the work she was doing, and they're preparing a legal claim, on the back of that. in many ways, Jeff, this is something we've seen a number of times before. There have been too many stories, I'd say, in which a content moderator whose, job it is to, you know, ensure that our feeds and our, our social media products that we use every day are, are safe and abide by the kind of policies that we all know about. And we're still, unfortunately seeing the psychological and traumatic effects of doing that work even as, in a kind of different corner of the internet, people are heralding the advent of AGI and this kind of new AI dawn. it's kind of chilling to me that we still have people who are doing this work, often without commensurate pay or wellbeing support and who are kind of suffering trauma. Now, there's almost certainly gonna be other elements at play here, but the fact that her family have kind of linked the two is very notable and, does follow a trend of, of people who have suffered in this way.

Jeff Horwitz 35:25

it's understandable, right? I mean, this, this issue goes back to, like, Casey Newton, I wanna say 2019. I was just starting out on "The Beat" when he was, racking up some really good stories on, on this subject. I mean, it's kind of like, the, Ursula Le Guin, Omelas," "The Ones Who Walk Away," story, except, instead of a child being tortured to guarantee the happiness of everyone, one, everyone's not happy, and two, at least it's an adult and they're being paid. But, like, This stuff is, just so awful. I can hardly imagine, doing that day in and day out as a shift. Like, I am protected to some degree by a sense that, like, when I end up being exposed to really awful stuff, I feel like it's for a higher purpose. It's not just, like, maintaining a status quo. If anything, it's, the opposite. but man, on, on the child safety stuff, like when I was, you know, reviewing content that Meta was serving to people who follow, who just follow gymnasts, don't do anything else, just follow some child gymnasts, my wife would wake me up in the middle of the night because I was screaming or, like, crying, uh, in my sleep, and that's, like, not something that is normal for me. and so, like, I, I, I do think that there likely are going to be gains from really good LLM-based classifiers. I do think it's underappreciated how significant some of those tools would likely be for the moderation business and already are inside some of the big companies, and people are a little reluctant to talk about that to some degree because it raises the question of maybe why things aren't getting better faster. But, to some degree, there's like, I don't see a world in which, we are not, basically paying some people to consume stuff that, just really, is never going to leave your head because it's so awful. I mean, I, very much understand how people who do this for a living, could end up with PTSD within days.

Ben Whitelaw 37:18

Yeah. I think to be honest that the BPO industry is still something that's, that's undercovered in the kind of social media space, right? Is there was a period where, as you say, Casey and others did a really kind of forensic look in it, but it's still an essential part

Jeff Horwitz 37:32

and then they just moved it... to, India and Kenya and, places where bluntly, people weren't gonna file lawsuits when the psychological impacts became apparent

Ben Whitelaw 37:41

Yeah. and also where, you know, the work has been kind of recategorized as, data labeling or model, labeling and, and the work is the same but, in a different coat basically. And, there has been some fight back and, and I'm, I'm glad to see that in parts of Kenya and Germany and Turkey and other places. And it's being joined up in ways that is, I think, uh, better for not only the people doing that work, but also in terms of kind of bringing about change. But it's not. it's slow, hard work, and we should kind of recognize that, I think, as an industry.

Jeff Horwitz 38:10

Yeah, I think-- I mean, I do think that, like, what I was saying earlier, just that one of the things that helped me when doing it was a sense that things might be changing, that it wasn't just treading water. I think that, the bleakness, some of the bleakness I think just comes out of like, I will be watching child abuse today, and I will be watching child abuse tomorrow, and I will be watching child abuse until I leave this job, because the child abuse will never stop, just so impossibly grim. I, I have a hard time getting my head around it.

Ben Whitelaw 38:37

Yeah. the faster that can change, the better, through improved labor rights and also AI moderation. that's my line on it. Geoff, we wanna quickly mention a story that we both like the look of but we don't really have time to go into, which is this, story that Anthropic is giving third-party auditors employee-like access to audit its AI. this comes on the back of a whole tranche of, whatever you call rogue attacks, breakout incidents. there's various different ways of referring to it, not all of them necessarily very accurate. But, do you wanna say something very briefly on, on, that?

Jeff Horwitz 39:07

yeah. this is obviously good. and I mean, I think it's something that with social media companies, researchers have never really had that privilege again, I'm gonna take Anthropic at its word here that it will be unencumbered access. My question is what happens if the researchers want to talk about things that they are seeing that are really problematic, what the limitations of that are? Because There are going to be a lot of devils and a lot of details. and, I think the history with Meta here is illustrative, which is that the company did, I think, head into doing some research partnerships, in good faith, with, uh, you know, Social science one was the first one. There was an elections partnership. There's been other stuff. I think it is, People have good intentions going in perhaps, but it turns out that letting other people rummage through your stuff and, criticize you publicly is, um, not something that everyone in the organization is perhaps gonna be fully comfortable with. And, and so I, I guess I'm, like, kinda counting down until some issues arise, even if the intentions are great.

Ben Whitelaw 40:05

Yeah,

Jeff Horwitz 40:06

and even if this is, like, a, thing that should be celebrated as a step forward and a good transparency thing, it, it absolutely should

Ben Whitelaw 40:12

Yeah. Yeah, exactly. The cynical journalist in us need to, uh, take a

Jeff Horwitz 40:16

Yeah. But, but it's gonna, but it's gonna be a problem, right? I mean, like, there's, there's, there's no way that, like, this is a stable thing where six years from now, like, there's, like, a steady team of people who are just, saying whatever the hell they want while having full access to the company's data. Like, just that's, that's not gonna be the way it works out, right? Okay. Yeah.

Ben Whitelaw 40:34

We can dream. We can dream. Um, so we're gonna transition now, Jeff, to our, Patreon segment for our paying supporters. listeners, if you liked today's episode, you want a bit more of Jeff and I talking about a story that, is eq- equally important in the platform governance world, then join us for that. We're gonna be talking about TikTok and some very kind of revelatory details about its child safety approach. So we'll dive into that now

Announcer 40:58

That's the end of this week's free edition of Ctrl Alt Speech. To hear the extended episode with additional coverage, become a Patreon supporter at patreon.com/ctrlaltspeech. That's C-T-R-L Alt Speech. If your organization is interested in sponsoring the podcast, contact us via our website at ctrlaltspeech.com. Thanks for listening