AI Breaches Add to Safety Fears as Trump Meets Anthropic CEO

0:02 Bloomberg Audio Studios. Podcasts. Radio.

0:06 News.

0:08 You're listening to Bloomberg Business Week with Carol Masser and Tim Stenevek on Bloomberg Radio. Okay, so we talked about some areas of the market that are getting our attention with, you know, the macro. This is still a macro story because AI has gotten so big, even though it's like company by company.

0:28 Yeah.

0:28 Yeah.

0:29 The sheer scale of this stuff makes it macro. So, okay, let's just go through some of this stuff, okay?

0:34 The first one just kind of blows my mind. You're going to do that first?

0:37 Dario?

0:37 Yeah. Yeah.

0:38 So Dario Amadei supposedly dining with President Trump last night. Here's the thing. Last night. But here's the thing. Joe and Kayleigh brought this up on the program.

0:45 Yeah.

0:45 I'm so grateful that I can hear their entire show before I start.

0:49 I know, I know.

0:49 Because I learned things like the president did not get back to Washington until after 9 p.m. last night.

0:55 Right.

0:56 So, like, who eats dinner? Is it dessert? Is it, like, I don't know.

0:59 I don't know.

1:00 Okay.

1:01 They order in McDonald's?

1:02 Oh, that's actually not a bad idea.

1:04 I know.

1:05 NVIDIA introduced a new double-layered AI security system that it says would have prevented the recent high-profile breach of Hugging Face by OpenAI's AI models. You were talking about this earlier.

1:14 Yeah, meta-platforms tapping. MongoDB, we've been talking about this. MongoDB is president and CEO to lead a new AI platform for enterprise customers that meta-CEO Mark Zuckerberg called, quote, the next major pillar of our business. So are they going corporate on us?

1:29 I don't know. That's a good question for Seth Figerman. He is the AI editor at Bloomberg News. He joins us here in the Bloomberg Interactive Brokers Studio. There's more headlines on AI that we didn't even get to with this.

1:42 Every day.

1:43 But I'm just wondering, from the standpoint of our audience and the markets, of the things that have happened over the weekend, what is the single most important development when it comes to your beef?

1:55 Yeah, I think the scale and number of the potential security incidences and misfirings of the AI agents that OpenAI kind of disclosed heading into the weekend is really top of mind.

2:07 Which I did not even mention.

2:09 Yeah, no fault of yours, too many headlines.

2:11 But the news crossed on Friday. They put out a blog post about it. We covered it here on our show on Friday.

2:16 Multiple places, right?

2:17 Multiple governments? Yes. And again, to be clear, these are not all equal things. Just because it tries to access or view content on a website doesn't mean that it's actually breached it or hacked it. But I think it speaks to just how long and how much and how often these products under testing have been sort of misfiring and acting in unpredictable ways. And so that's sort of the context I would look at stuff like the Dario Amadei meeting with Trump and the NVIDIA announcement that you mentioned.

2:41 Do we have anything on the... We were talking about this this morning in the newsroom. We're like, wait, is this... We didn't hear anything about the dinner. So all we heard was the dinner was supposed to happen.

2:49 It's our understanding that it did happen.

2:51 Okay.

2:51 And to your point, I think it happened at like 10 o'clock last night, which as a parent of little kids makes me feel kind of like, I don't know when I can do that. We haven't gotten much of a readout yet. I don't know if Trump has talked about it recently in any of his public... Public Q & A sessions, but not much is coming from the anthropic side.

3:06 Is that kind of weird that we're not hearing anything?

3:08 I wouldn't expect much immediately from the anthropic side. I think you kind of let Trump be Trump and then share what he wants to share in some of these Q & As. But our understanding heading into that meeting was that it came in response to Dario's calls recently to potentially slow down the pace of development, which we know from his own public rhetoric, Trump is not totally a fan of. And at the same time, there's supposed to be another meeting tomorrow with other executives, including Trump. Dario, Trump, and other folks on the Hill to kind of talk generally about AI safety and pacing.

3:35 Well, it also comes off of the meeting with President Xi, who actually did seem to talk about global coordination on this, correct?

3:42 Yeah, and I think that China, for all of the narrative that we have in D.C. right now, China has been pretty tough in regulating its companies. It does really take a lot of oversight over how its firms handle AI. You could argue the objectives and what it's fighting for might be a little bit different than here, but I don't think any of these countries want just unfettered AI technology.

4:00 development and AI running them up.

4:02 Why don't these guys just put garbage? Why don't they just do it if they're testing these systems?

4:06 Well, I don't know if you saw the SNL skit over the weekend about Dario, but I think that does kind of capture a certain amount of public sentiment here. Yes, yes, yes. There is that here. Like, why are you begging people to stop you from doing the thing that you could maybe do yourself?

4:16 Just do it.

4:16 But it's the competitive landscape. I think what Anthropic and Dario might say is they feel immense pressure simultaneously to safeguard the tech, but also to be at the vanguard of that tech to really set the standards.

4:26 Are they right, Seth?

4:28 No, that's above my pay grade.

4:29 Okay, fair.

4:30 Yeah.

4:31 One thing that I don't think is above your pay grade is when it comes to productivity and the software, you had a great, we had a great conversation on Friday about the way Microsoft is sort of going back to like the days, you know, of, I don't know, bringing productivity with AI, which I think is not surprising to a lot of people who've been following the Microsoft story for years. I think what is more surprising is what we were talking about last hour, about how meta is bringing in this meta enterprise platform. And, and, you know, we had a similar conversation when we were talking about meta a few months ago, working on something like this.

5:12 You know, it's the idea that, okay, we kind of use these, I'm not going to call them tools. Like Instagram is like such a time suck for people.

5:21 But people who sell stuff, use it as a tool.

5:23 That's a good point.

5:23 Fair.

5:24 And they do sell stuff.

5:25 Fair point.

5:25 Yeah.

5:26 Enterprise platform for large companies is, is this, this is, It's so far removed from what Meta has done over the last 20 years.

5:31 Yeah, I mean, I think there was a joke at one time that Meta was spending hundreds of billions of dollars to build AI so that it could give you a Snoop Dogg-themed chatbot. Was that really the best use of its investment? And it seems like even they have acknowledged that we need to be a little bit more thoughtful and intentional about that and show customer adoption, but also prove to Wall Street that there's money to be made here. And so I think that's happening across a couple different vectors there. they have really done a good job compared to some other folks in securing more computing resources.

6:00 So suddenly you have ample computing power, not just to build the tech, but potentially to supply to other third-party partners and customers. At the same time, they have caught fire a bit in the last week or so with that Muse agent, which it's not so much like it's like introducing tech that didn't exist before. I think it just kind of puts like a nice intuitive wrapper around it and connects to products that we use every day.

6:20 And to the extent that you're comfortable with Meta having access to your Gmail, your contacts. Absolutely not. Yeah, but that's part of it.

6:28 I mean, we didn't get a chance to talk to you about this last week. We had a great chat with Dave Lee about this. But there was this New York Times sort of review of Muse, and a couple of friends sent it to me, and one of them was like, wow, this is a great review. I read it. And the guy who did the story connected his bank account or maybe a bank account that he created for this.

6:47 Like that takes a really extreme level of trust with an agent when all this chatter is happening with like these rogue sort of agents going off and doing their own thing.

6:55 Well, two things. One, I also played with Muse over the weekend and I had it like work on my behalf to like get me a better insurance quote and help me change my car insurance.

7:02 Did it work?

7:03 It did, but I had to give it access to my Gmail. I had to create a form of payment via Stripe as an intermediary to do that. And someone who, you know, it tested my comfort levels. I don't know that I'm going to let it keep having that access indefinitely. But to your other point, these two things are very much happening side by side at the same time that these companies are warning. about AI safety risks. They're trying to really accelerate adoption of these agentic products into our everyday lives.

7:28 Because what is to stop an agent from saying, hey, I want to go shopping and I want to send it to my card to you.

7:35 Hold on.

7:36 Sorry, honey.

7:36 No.

7:37 It was the agent. Exactly. Exactly.

7:40 But I mean, right?

7:41 Yeah. I think Mark Zuckerberg has kind of said this more clearly than some other folks to his credit. He has said, he personally feels like we don't need necessarily like the regulatory oversight body or whatever. Like these businesses have their own natural business.

7:52 And they do so well.

7:54 Yeah, absolutely. So the flip side is mega has a long history of privacy issues.

7:59 Okay.

7:59 So maybe, maybe this is an opportunity for a new startup. The startup instinct has raised a billion dollars.

8:06 I thought you were going to pitch.

8:09 You're so used to pitches. No, that's funny. Uh, I'm just using it as a segue to get to this story. Um, that was written by another member of the team. Actually, you were one of the editors on this too. AI agent startup instinct raised a billion dollars at $ 10 billion valuation. The reason I brought this one up, it wasn't even a part of the plan today, but every time I log into X, like somebody, some tech person is talking about this amazing thing that instinct was able to do for them.

8:32 Yeah. I think they're all part of the same frenzy. It's like, how do we translate AI tech into something you can use every day to act on your behalf? The Instinct thing, though, is interesting. I can stop in a minute. But I don't know if this thing will be a big thing in a year or two years or two weeks. But it costs so much more money for a startup to operate a product like that than it would have five years ago for kind of like a flavor of the week app to launch. And so they just have to raise an inordinate amount of money to meet that demand. It's not like they're sharing photos. Yeah.

9:02 And so I think part of why you're seeing, oh, my God, this company I hadn't heard of two months ago valued at $ 10 billion, they just need to raise hundreds of millions of dollars to cater to the demand that they're seeing for a product that's always on running in the background. And so I think we'll see more of that.

9:16 I think it was our own Mandeep Singh who talked about Instinct with me because I wrote it down and I was going to take a look at it.

9:21 People love it.

9:21 You need an invite.

9:23 You're on the invite list.

9:26 I'm trying to think where else we should go.

9:27 I mean, I don't know. I don't know how you guys keep sense of it all.

9:31 We don't.

9:31 We can get an agent for it.

9:34 We had two professors last week and we were talking about disruption and making the point that there are companies that we take for granted that will be with us forever. Think about the Mag7. Their argument is that they may not be around in 10 years or 20 years. We've seen that, certainly these tech cycles and these innovation disruptions happening faster.

9:57 Do we still not know necessarily who's going to be the leaders? Is it safe to say that?

10:01 I think we don't know who the leaders are going to be. I think we know that the firms at the vanguard of AI development have stayed pretty consistent the last six months or so. Like Anthropic, OpenAI, they trade places depending on the week. But they've been at the vanguard. Meta fell out. Now it's kind of getting close again to the frontier. There's that kind of swapping that happens.

10:20 But when you think about stepping back, it's not just how you build the technology, but what you build with it, how it gets used for people. Part of why Muse is taking off, again, is putting it in a more intuitive interface. It's not necessarily creating new technology. If you believe that six months from now, everyone's going to have their personal agent or agents who are streamlining all sorts of tasks, then yeah, there are definitely going to be products and services that you might have used previously. that you just don't need to anymore because there's a little bit less friction. I don't know that we're there yet, but that's, again, that gets back to the SaaSpocalypse kind of concerns.

10:48 All I would say is that people like things to be easy, and if you start to do it and it works, I.

10:52 Think people are going to be all in. I don't feel totally comfortable with it either.

10:54 I mean, you just described why Apple has succeeded so well with hardware. They make it easy.

10:59 Yeah, exactly. Great. We went all over and so appreciate it.

11:04 Seth is the best.

11:05 You're like amazing.

11:06 Seth Fiegerman.

11:08 He is AI editor at Bloomberg News. Joining us right here at our Bloomberg Interactive Brokers Studio.