Meta to Track Employee Keystrokes and Clicks to Train AI | Bytes: Week in Review | Marketplace Tech

0:01 [music]

0:06 Welcome to Marketplace Tech Bites, our weekly review of a few

0:09 of the biggest stories making headlines across the industry.

0:11 [music] I'm Stephanie Hughes.

0:13 This week, Apple CEO Tim Cook to step down,

0:17 Meta to capture [music] employees mouse movements for AI training,

0:22 and Roblox [music] settles with states for $36 million over child safety issues.

0:27 I'm joined by our regular contributor Anita Ramaswamy,

0:30 a columnist at The [music] Information.

0:31 Hey Anita.

0:32 Hey Stephanie.

0:34 So, we are going to start with our bite of the week.

0:35 This is one number that gives you a hint

0:37 about something that happened this week in tech.

0:40 Anita, what is your number?

0:42 My number is 3 billion, and that's the number of iPhones,

0:46 or at least the number of iPhones that Apple has sold since 2011,

0:50 which is when Tim Cook became CEO of the company.

0:53 iPhones, a major money maker for them.

0:55 Okay, so let's talk about this.

0:56 So, after 15 years on the job,

0:58 Apple announced this week that CEO Tim Cook is moving on from the role.

1:01 He'll become the company's executive chairman,

1:04 and he's being replaced by John Ternus,

1:06 who is currently the senior vice president of hardware engineering at Apple.

1:10 Anita, Cook's been in this job, as you mentioned, since 2011,

1:13 which seems like a long time in both actual years,

1:16 and even longer in like tech years.

1:18 What was your reaction when you first heard this news?

1:21 Yeah, my reaction when I first heard this news,

1:23 this is obviously a massive shift for the company,

1:25 but it's not totally unexpected.

1:27 We had seen a lot of reporting kind of talking

1:29 about Tim Cook and how he was planning his successor,

1:32 and John Ternus was the most likely candidate.

1:34 I think what this shows is Apple is focusing and going all in on hardware.

1:38 They make a lot of profits from their software and services arm,

1:41 from their App Store and sales and transactions there,

1:44 but this really shows that what they're investing their capital behind

1:47 and what they're putting a lot of their strategic focus behind is possibly,

1:51 you know, introducing new devices or tweaking old ones,

1:53 and that that goes to the heart of what their bread and butter is,

1:57 which is iPhone sales, and that has been the case for, you know,

1:59 all of Tim Cook's tenure.

2:01 Yeah, what would what might you expect under Ternus in this role?

2:06 So, I think the thing with Ternus is, you know, he came up in Apple.

2:10 So, in 2014, Apple released the Apple Watch, you know, he was around for that.

2:14 He was running hardware engineering, you know,

2:17 for all the latest models of the iPhones, for the the MacBooks.

2:20 And so, I think what to expect from Ternus, hopefully,

2:23 and this is something I think investors will be,

2:26 you know, wishing for, that he does some product innovation.

2:29 Everyone talks about going taking Apple back to a place where

2:32 it's more similar to what it was like under Steve Jobs.

2:34 Tim Cook was considered to be a really good

2:37 supply chain expert and a really savvy business person,

2:39 and kind of helped the company manage

2:41 their costs and navigate through some tough economic headwinds,

2:45 but at the end of the day, John Ternus is an engineer,

2:47 and so that's going to be his focus and how he likely prioritizes things.

2:51 Yeah, he's a hardware guy.

2:53 Um, okay, so last month Tim Cook gave an interview to the Wall Street Journal,

2:57 where he was asked what advice he would give to his successor,

3:00 and this is what he said.

3:01 So, I would say, you know, be yourself,

3:04 keep a firm North Star on the values of the company.

3:11 Anita, what value, either one of Apple's values or values in general,

3:15 do you think Ternus should look to as he navigates both this transition

3:19 and also also this kind of wild moment in the tech sector?

3:24 It's a good question.

3:25 I mean, I think I'm not I'm not a CEO of a multi-billion dollar company,

3:29 but I'll give my two cents here,

3:30 which is that I think that the innovation is really important for Apple.

3:34 We've seen some stagnancy in iPhone sales

3:36 in the last several years, and, you know, with this whole AI race,

3:40 it remains to be seen whether consumers

3:42 like actually want AI features in their iPhones.

3:44 Yeah.

3:45 I think the Apple Intelligence rollout wasn't, you know, too well received,

3:48 but as people are using AI in so many different,

3:51 um, you know, other places, Apple is already kind of behind in that race.

3:55 So, I think where Apple really can differentiate itself is trying

3:59 to come up with new devices that people actually want to wear.

4:02 We saw a little bit of, you know,

4:03 mixed reaction to the Vision Pro, to their VR headset,

4:07 and I think it's just a matter of finding what is the next form factor,

4:10 what is the thing after the iPhone that people, you know, really want,

4:14 and they're going to show up in droves to go and buy.

4:17 Yeah, you know, under Cook's leadership,

4:19 Apple's annual revenue nearly quadrupled.

4:21 Last year, it was more than $416 billion.

4:25 There's no question that Tim Cook has

4:27 been enormously successful in helping Apple make money.

4:30 It'd be basically impossible to, you know, continue at the same pace.

4:33 What do you think success looks like for John Ternus?

4:38 I think again it comes back down to his engineering background,

4:41 and so that is not necessarily introducing new product lines altogether,

4:45 but maybe sort of tweaking existing products so

4:48 that they are more tailored to to just

4:51 at least continue the growth trajectory that Apple

4:53 has been on for so many years.

4:54 Of course, it's a massive company,

4:55 and you're totally right to point out that, you know,

4:58 accelerating growth from here is a really really tall order,

5:01 but I think at the same time Apple can't rest on its laurels, and, you know,

5:04 they need to continue to extend the market in some way, shape, or form.

5:08 And so, that's where it's going to be really exciting to see whether, you know,

5:11 Apple will pull ahead in terms of this race

5:14 to create the next computing interface that people want to use.

5:18 Okay, let's get to our next story.

5:20 Meta to capture employees mouse movements for AI training.

5:25 So, Reuters reported this week that Meta is

5:27 installing new tracking software on its US employees' computers,

5:30 which will capture mouse movements, clicks, and keystrokes.

5:34 This data will go to train the company's AI models,

5:36 and it will not be used for performance reviews.

5:39 Managers also won't be able to access it.

5:41 Meta said in a statement to us, quote, "If we're building agents to help

5:44 people complete everyday tasks using computers,

5:47 our models need real examples of how people actually use them." Anita,

5:52 what do you think of this approach that Meta's taking here?

5:55 Um, you know, on a personal level, it doesn't sit that well with me,

5:59 and that's just because the idea is that, you know,

6:02 this is employees who are going to be training AI employees to replace them.

6:06 Yeah, so it's a little it's a little on the nose, I thought.

6:08 Yeah, I was wondering about that.

6:09 You know, Meta is reportedly planning to cut

6:11 10% of its entire workforce next month.

6:13 That's 8,000 people.

6:15 You know, if you're working at the company now,

6:18 and it's watching everything you do, and training training the AI models,

6:23 do you just feel like you're training your replacement?

6:26 Yeah, exactly.

6:27 I think that's the concern here, and, you know, for right now, yeah,

6:30 the the data is not going to be used in performance reviews and all

6:33 of that, but I think we have seen a trend at large companies,

6:37 especially large tech companies,

6:38 where they're pushing employees to use AI more and more.

6:41 And so, we have seen a little bit of this race of, you know,

6:44 companies measuring how many tokens,

6:45 for example, that their employees are burning, you know,

6:48 asking AI, uh, questions or using AI for their work.

6:51 And that's something that Meta tried and then sort of pulled back on, um,

6:54 that we had reported about at The Information.

6:56 I just think that across the board this AI usage is being pushed,

6:59 and it is a new reality of the life the lives of many people at tech companies.

7:05 Why do you think the company's going

7:06 to its current employees for this, as opposed

7:08 to maybe hiring new people or contracting with new people to do it?

7:12 Yeah, so this is a really good question, Stephanie,

7:14 and I think this is, um, you know, a a really important one,

7:16 which is that Meta and all these other companies that are training massive AI

7:20 models are running out of data to train them on and to improve them.

7:24 So, we saw that Meta, you know, took a big stake in Scale AI,

7:28 and what Scale AI does is they are like a data labeling,

7:32 um, data gathering firm that helps Meta

7:34 get that raw material to train AI models.

7:37 And quickly, we're running out of material on the public

7:40 web that's available for free for these AI companies to use.

7:43 And so, the next layer is going to be like scraping

7:46 any source that can possibly be used to tweak the models,

7:49 and that's why they're turning to their own employees,

7:51 because now we've already used all the publicly available information.

7:55 So, I think this is Meta sort of getting creative of how can we how

7:58 can we tweak the models when there's not

8:00 that much input anymore that we can use.

8:03 That's so wild that we're running out of information.

8:05 Um, I want to [laughter] I want to talk about potential costs of this to Meta.

8:09 You know, I feel like employees may be more guarded as they do their work.

8:13 I certainly would be.

8:14 And it made me wonder, could this hurt creativity,

8:17 or might it just make workers more likely to leave?

8:21 Yeah, I was, um, seeing some of the discourse

8:23 on X about this, and there were some folks who were,

8:25 I get I think sort of tongue-in-cheek joking that, you know,

8:28 they they want to see how many people ended

8:30 up taking medical leave or something like that, you know,

8:32 during this period, because they don't want their work

8:34 to actually be used or to be scraped by Meta AI.

8:37 I don't know how big or how pronounced the backlash initially is going to be,

8:41 just because of the broader factors at play in the job market,

8:44 you know, it's a tough time economically.

8:46 Um, but I do think there will be

8:47 some real resistance and some real backlash to this.

8:50 And at least on the margin, you know, someone who has a lot of, you know,

8:53 job opportunities might not want to work at a company

8:55 like Meta if they're going to be surveilled.

8:58 At the same time, I think that whatever these big tech companies like Meta do,

9:02 the rest of the tech industry tends to follow.

9:04 So, I wouldn't be surprised if we saw similar initiatives from other companies.

9:08 Yeah, that was my last question here,

9:10 is how commonplace do you think this kind of monitoring is going to become,

9:13 either, you know, both at tech companies and then companies overall?

9:18 My feeling is that it's going to become ubiquitous.

9:21 I think, you know, when we see a company as large

9:23 as Meta doing this, that it it's going to become very prevalent.

9:26 And, you know, not every company is

9:28 really at the frontier of training AI models.

9:31 A lot of these smaller tech companies,

9:33 I could see a world in which they, you know,

9:35 use scrape the data from their employees and then maybe

9:37 sell it to a company that is training the AI models.

9:40 And especially when we talk about the future of AI being

9:43 with agents and the sort of agentic workflow that, you know,

9:46 the the AI models going to take actions

9:49 on your behalf and really function like a human,

9:51 I think that the mouse clicks of an employee are

9:54 tremendously valuable data points to sort of train AI agents.

9:58 So, I just see this as an inevitability and I

10:01 think different companies will implement it to varying degrees.

10:05 All right.

10:06 Let's get to our final story.

10:08 Roblox settles with states for $36 million over child safety issues.

10:13 So, the popular the popular video game company Roblox

10:15 has reached settlements with the attorneys general in Alabama,

10:18 Nevada, and West Virginia.

10:20 Roblox says it's going to work with the states

10:21 to make it safer for kids to be online.

10:24 A number of cases have been filed in both state and federal courts against

10:27 Roblox saying the company has failed to protect

10:30 minors on its platforms from predatory behavior.

10:33 Anita, what do these settlements say to you?

10:37 These settlements are, you know,

10:38 it's coming to a head this issue that Roblox has had for a long time.

10:43 The majority of their users are under 18 and this is a really thorny issue

10:47 that Roblox is not the only company that has had to deal with this sort

10:51 of thing and I think that it shows that age verification laws and procedures

10:56 and processes are just becoming more

10:57 and more prevalent throughout tech as concerned parents,

11:01 concerned families make uh an uproar over some of the content

11:05 that their children have been exposed to on the platform.

11:07 Yeah.

11:08 Um I talked earlier today with Matt Kaufman

11:10 who's the chief safety officer at Roblox and he

11:13 says the company is working with the states

11:15 to set the standard for kids interactions online.

11:18 And he highlighted this age verification tool

11:21 that the company rolled out last year.

11:22 It's just a snapshot of your face and it just from that snapshot

11:26 you can guess somebody's age to a pretty high degree of accuracy.

11:30 It's within about 1.4 years if you're under 18 years old.

11:34 And so, the idea is this is this will

11:36 help verify how old their users actually are.

11:38 You have to be at least nine years old in order

11:40 to communicate with other users your own age on Roblox.

11:44 What do you think this tool will accomplish?

11:47 I think for them it's uh you know, somewhat of a response to the regulatory

11:52 and legal issues that they've been facing.

11:54 You know, they settled with Alabama, they settled with West Virginia.

11:57 We also saw them settle earlier a similar

12:00 matter in Nevada where they had, you know,

12:02 paid $10 million state and so I think for Roblox this is, you know, a necessity.

12:07 They have to do something otherwise they're going to keep

12:10 facing this sort of pressure and I think it's

12:12 also worth noting that there are other companies that have

12:15 implemented this sort of technology a little bit earlier.

12:18 Meta has, you know, rolled this out across some

12:20 of its products I believe around 2022 time frame.

12:23 Um OnlyFans which is another, you know, popular creator platform they rolled out

12:28 age verification based on face scanning which

12:31 is what Roblox sounds like they're they're doing right now um a couple

12:34 years ago and so I think it was just a matter of time

12:36 for this sort of thing got adopted at a company like Roblox.

12:40 Yeah, you know, to your point Roblox the company is over two decades old and I

12:44 also asked Kaufman why this age verification is

12:47 something that the company is just now doing.

12:50 It takes a long time for the the technology to mature and for it to be tested.

12:55 And so, when we think about something like

12:57 age checks and using a face to age estimation,

13:01 we implemented that as soon as we felt that the technology could scale

13:05 and was accurate enough to really work with our large audience which is global.

13:11 Anita, what do you make of this?

13:12 Look, I haven't drilled deep into exactly

13:15 how mature the technology is or, you know, what constitutes it being ready,

13:19 but I do think that this is a problem

13:21 that maybe Roblox should have addressed a lot earlier.

13:24 Uh you know, investors have not been

13:25 reacting well to all of these issues naturally.

13:28 Their stock is down more than 30% in 2026 so far and they've been

13:32 on this push to increase the amount the share

13:35 of users that are over 18 years old, but I think um you know,

13:38 this is an additional safety measure that that could have been

13:40 implemented given that other tech companies implemented a little bit earlier.

13:45 Roblox is, as you mentioned, not the only company dealing with lawsuits

13:49 over how kids interact on its platform.

13:51 It's also a major issue for social media in general.

13:54 What sort of protections writ large do you think or hope

13:58 we could see for kids in the future as they go online?

14:02 I think it's a really nuanced issue

14:03 that kind of applies to each uh specific platform,

14:06 but at the end of the day you have to verify that users are

14:09 of an age that's legal and that they're

14:11 not exposed to harmful content if they're under 18.

14:14 And so, whatever, you know, technology can be used to make that happen

14:18 should be rolled out as soon as possible.

14:21 And I think, you know,

14:22 there are other there are other measures that companies can take to also

14:25 try and shift that mix like not marketing directly to children or, you know,

14:29 I I think there are there other protective measures that can can occur there.

14:33 Roblox has done some of those things and that's part of why their users over

14:36 18 have been growing a lot faster of a rate than their users under 18,

14:41 but there's still a long ways to go and I think we're going to see

14:43 this play out at a host of tech companies over the next several years.

14:48 Anita Ramaswamy, she is a columnist at The Information.

14:51 Anita, thank you so much for [music] talking with me.

14:53 Thank you.

14:54 And thanks to all of you for joining

14:56 us for Marketplace Tech Bites week in review.

14:58 Daniel Shen produced this episode.

15:00 I'm Stephanie Hughes.

15:14 This is APM.