Anthropic vs Pentagon & OpenAI’s Deal, Apple Discusses Google Hosting Siri, Supercharged Power Lines

0:12Aaron Tilley:Welcome everyone to The Information's TI TV. My name is Akash Pasricha. It is Monday, March 2nd. Today on the show, we are unpacking what we know about Anthropix's plans to sue the government over its supply chain risk designation and OpenAI's deal with the Pentagon. on. We'll then bring on an OpenAI investor from Coastal Ventures to hear his view on all of this and also hear a little bit about how he is thinking about the company's roadmap at large. We're then turning to an exclusive story from our Apple reporter published today about how Apple could increasingly turn to Google for cloud support as its own cloud efforts have struggled and it looks once again to outside partners.

0:53Aaron Tilley:And finally, we will end the show with a look at how the AI data center build out has spurred the acceleration of extra high voltage power lines. That is the subject of our AI infrastructure newsletter out today. It's going to be a busy show, so let's get right on into it. Anthropic is readying its court challenge to the government's designation that it is a supply chain risk after the company wouldn't budge on a contract with the Department of Defense. Meanwhile, OpenAI has struck its own agreement with the Pentagon on. For more analysis on how both of these dynamics could play out, I want to bring on Erin Wu, who has been covering this story very closely over the past week.

1:32Aaron Tilley:Erin, welcome back to the show. It's great to have you here.

1:34Erin Woo:Hey, thank you so much for having me.

1:36Aaron Tilley:Okay, so I know that a lot has happened over the weekend, and a lot happened in the days leading up to the weekend as well. A lot has happened, okay? There's a lot going on. Walk us through just the chronology of the past week and then where we sit today with these dynamics? Yeah.

1:53Erin Woo:So essentially we've had a few weeks of escalating negotiations between Anthropic and the Pentagon over essentially what Anthropic will allow the Pentagon to use its AI for. So the Pentagon wants to be able to use AI companies' AI for, quote, all lawful uses. And Anthropic has two red lines, no use of its AI for mass domestic surveillance and no use of its AI for fully autonomous weapon systems. So what happened last week after these weeks of negotiations that have spilled out into the public, Anthropic CEO Dario Amadei and Pete Hegseth meet face-to-face continuing negotiations. Anthropic after this, days after this, publishes a statement essentially saying that it's going to hold to these red lines.

2:39Erin Woo:that on Friday, they have a deadline to essentially either agree to the Anthrop... On Friday, they have a deadline.

2:50Aaron Tilley:It's been a long weekend. I know you've been working the phones all weekend. So let me help you with this time. So Friday, they have the deadline. 5.01 p.m. is when the Department of Defense and Anthropic have to reach a deal. Anthropic has to say whether or not they will budge on these red lines, essentially. And then where we sort of left off last week is OpenAI came into the mix. How did they come into the mix?

3:20Erin Woo:Yeah, so OpenAI has been negotiating this deal, essentially in parallel as Anthropic has been negotiating its deal. And so at 5.01 p.m. ET, the deal does not come together for Anthropic. and Pete Hegseth declares Anthropica supply chain risk. Later that night, OpenAI announces that it's signed a similar but different and many important aspects deal with the Pentagon to provide OpenAI's models on classified systems.

3:49Aaron Tilley:Okay, so we have the OpenAI agreement and now we have the Anthropic supply chain risk designation. I want to get into both of these pieces separately. was open ai's agreement with the pentagon any different than what supply what entropic was

4:08Erin Woo:sort of fighting against yeah yeah so in many important ways yes entropic was fighting against this contractual language that said that the pentagon could use its ai for any lawful uses open ai agreed to this contract language however open ai says that it negotiated other bits of the agreement that it's hoping will help it keep the exact same red lines as Anthropic, which is prohibitions against the use of its AI for lethal autonomous weapons and mass surveillance. So OpenAI says that because it can keep its safety guardrails in its models, it can use technological means to prevent the Pentagon from doing those things.

4:49Erin Woo:OpenAI also negotiated its agreement so that it only works in the cloud, not on edge deployed models, which OpenAI says means that it can't be used on lethal autonomous weapons. And finally, OpenAI negotiated the right to have OpenAI engineers cleared personnel working with the Pentagon to help implement it, which they're hoping will give them more visibility into what's actually going on.

5:12Aaron Tilley:So in other words, OpenAI is saying that, hey, what we may have given up a little bit in terms of language, we actually have those extra guardrails in place on the technology side. And so we can ensure that we have a little bit of control over what the technology is being used for.

5:28Erin Woo:And just to note, we've seen part of the contract, which in fairness is more than we've seen for Anthropic. So OpenAI released a couple paragraphs of contract language over the weekend that deal specifically with the lethal autonomous weapons and mass domestic surveillance causes. So we've seen two paragraphs of the contract, but we haven't seen the whole thing. And so we don't know how it works with safety guard or for deployed engineers or whatever. And so there's still a lot of unanswered questions. Okay.

5:59Aaron Tilley:Now, how have people been reacting to this? Because over the weekend, I mean, I've seen people posting screenshots of Anthropic climbing up the leaderboard. I've seen people with very strong reactions both ways to OpenAI's contract. Just walk us through how people are thinking about these two. Right.

6:18Erin Woo:So it's been a huge weekend for sitting on Twitter and watching people post about it. Over the weekend, Sam Altman and some of his top execs did an Ask Me Anything on Twitter. So huge weekend for people posting. I mean, overall, there's a moment when the deal was first announced, there was a moment when people were like, oh my God, how did OpenAI get the exact same deal that Anthropic failed to get. But over the weekend, that sentiment really swung towards OpenAI seems to have accepted something that Anthropic had concerns about. And so that raises a lot of questions about, for example, how OpenAI employees feel about this.

6:55Erin Woo:As of last night, around 100 of them had signed this pledge, co-organized with Google around standing in solidarity with Anthropic. And so OpenAI employees, especially researchers, are in this incredibly hot talent market. We all remember the AI talent wars from last summer. And so if OpenAI employees decide to kind of try to speak out against this, there could be some real impact there. On the other hand, we don't totally know how this contract works. And OpenAI has said that it kept the same red lines as Anthropic. And so it's possible that OpenAI employees are happy with this contract.

7:32Aaron Tilley:Now, what about the Anthropic lawsuit? I think as far as we know, Anthropic has released a statement, they said, we plan to challenge this designation in court. So we don't know a lot, but I wonder if you've talked to any lawyers or any experts around what sort of legal argument they might use and how much of a chance you think they have to overturn this designation?

7:57Erin Woo:Right. So essentially what Pete Hegseth has done is declare anthropic a supply chain risk. And in his tweet on Friday, he used a really expansive definition of supply chain risk. And so what he wrote was that he was barring Anthropik from having any commercial relationship with anyone who does business with the Pentagon, which is so many companies. For example, Anthropik's cloud providers, Amazon and Google, are both military contractors. And so theoretically, this could cut Anthropik off from not even just being able to sell to customers, but being able to receive cloud services. And so this is a really big deal if courts uphold it, as Pete Hegseth has talked about.

8:37Erin Woo:But I think the question is, would courts uphold it? My colleague Rocket talked to some lawyers on Friday. This idea is that, A, this is a law that was designed for foreign companies. Like, it's not designed to be applied to an American company. So there's that. But Dario, the CEO of Anthropic, is also arguing that it would only ban Anthropic from being used much more narrowly in the military supply chain. So there's that argument secondarily. However, there's also the point that ports may be more likely to defer to the federal government on matters of national security. That was also a point that came up in the TikTok ban, if you remember that.

9:20Erin Woo:And in practicality, companies who are dealing with anthropic may just find it easier and safer to cut ties.

9:29Aaron Tilley:Last question for you, Erin. You mentioned the TikTok ban. You have covered social media extensively. Yes. Now you're covering the way that AI companies are interacting with the government. Do you see any similarities, differences in terms of how this is all playing out? I mean, these are issues that have spanned multiple administrations, right? And so I just wonder how you're thinking about, is this a story that we've seen before? The supply chain risk is obviously something new. Can we learn anything from past Silicon Valley brush-ups with the government, I guess?

10:07Erin Woo:Yeah, I mean, I think this is new in a lot of ways. And there's also a lot of history that comes into it. So, I mean, like if you look at tech companies working with the military, there's obviously like a rich history of that happening and also employees protesting it. So, most famously, you have Project Maven in 2018, like Google declined to renew its contract after three, after 3000 or thousands of Google employees signed an open letter opposing it. So, there's history of tech companies working with the military and companies response to that. Obviously, in a lot of ways with AI, we're in very uncharted territory.

10:42Erin Woo:And look, when you're asking that question, what I was thinking, I was talking to a lawyer yesterday about these two questions of lethal autonomous weapons and mass surveillance. And I was asking like, oh, like, what can AI do with that? And what this person was really hammering home for me was that like Congress needs to act. And in that way, we're in a very similar situation to social media where we really don't have any meaningful regulation of social media on a national level, despite these services being around for years and years. So I think if anything, one of the real parallels that we have is just tech companies being so far ahead of Congress in terms of regulation that we're now in this gray area where, yeah, private companies are trying to decide what should be acceptable for this kind of technology.

11:30Aaron Tilley:Great. Well, Aaron, I want to thank you for coming on. That is Aaron Wu, our reporter here at The Information. For more coverage on OpenAI's agreement with the Department of Defense, I want to bring on Ethan Choi, who focuses on growth stage investing at Kostla Ventures. Kostla, of course, is an early investor in OpenAI. Ethan, welcome to TITV. It's great to have you here. Great to have you here. Thanks so much for having me. So you work with OpenAI as an investor. What is your view of the agreement with the Department and defense and how things have played out.

12:00Akash Pasricha:Look, this isn't really, you know, these two companies obviously have been battling it out on the model front, and this really isn't about the model this time. It's really about contract language. And the fight, you know, here is mostly about one phrase, actually, which is the phrase of all lawful purposes. And, you know, ultimately comes down to the values and the philosophies of both companies. And you're seeing that play out in real time.

12:30Aaron Tilley:Okay. And so do you see, people are concerned that OpenAI has said that they're okay with that phrase and they're saying that they have other guardrails to sort of, you know, protect against uses that they don't approve of. I mean, do you understand the concern from people? You know, walk us through how you're thinking about that. Yeah, look, you know, Sam did an AMA

12:54Akash Pasricha:And he broke it down really well, which is, you know, at the heart of this whole debate is one meta question. And one maybe that's more important than the red lines, which I can get into. The meta question is, does a private company, even as important as Anthropical or OpenAI, do they have more power to decide how the government operates than democratically elected officials in the government? And I think that's the heart of the matter here, because ultimately we have a government that has a Department of War that has the responsibility and the context and the people and the expertise to interact with the technology to fulfill its purposes.

13:48Akash Pasricha:And Sam has ultimately decided that it's not up to OpenAI as a company and their leadership to determine how to use that. And I think that's the right point of view. I don't think there should be a single company, private company, that can control what the government specifically does when it comes to matters of national defense.

14:09Aaron Tilley:Let me ask you about this supply chain risk designation. as an investor who's investing in the AI sector at large, you know, and we'll step outside of the open AI sandbox for just a minute. I mean, does that designation or the idea that the government may grant or label a company with that designation, does that cast any kind of a chill for you over investing in AI? Or does it add any kind of a risk for you as an investor?

14:40Akash Pasricha:Not on the broader spectrum of things. And, you know, it is unprecedented, just to be clear, that the government is looking at applying that kind of designation to a private company. You know, with that said, I would imagine in defense tech that if there are situations where a company decides not to provide the technology or the capabilities that the government needs to fulfill its responsibilities and activities, then it seems fair that the government has some repercussions and abilities and teeth to enforce the rights that they have, which is as a government. and using a specific vendor, they have the right to protect our nation's citizens.

15:30Akash Pasricha:So I can understand it to a large degree, but I also understand the concerns that others may have where if the government becomes very heavy handed in that manner, then what rights do we have as citizens to protect ourselves?

15:45Aaron Tilley:And one more question for you on this note here. I am curious around the differences in the technical capabilities between OpenAI and Anthropic. I mean, is it your understanding that there is a difference in how much control OpenAI has over its technology compared to Anthropic? You know, I'm sort of trying to understand the argument here that the technology guardrails could really uphold the risk against the language, I guess.

16:13Akash Pasricha:Yeah, that's right. Right. So maybe coming back to my first point where this is actually a battle between contract language and not actually the models. If you look at, you know, the difference between OpenAI and Anthropic, there are small differences in the actual deployment of the technology. OpenAI largely has focused on enabling its application in the cloud. Anthropic through Cloud Code and their plugins actually is tied deeply into existing applications that we use every day. Things like Microsoft Word, Microsoft Excel. And there's a slight difference in that deployment, but it's again back to the contract language.

16:54Akash Pasricha:And what's happened here is Anthropic has decided that they want very, very bright red lines and enforceable contracts to try and control and narrow the permissions up front. And what OpenAI has decided to do is actually live under the umbrella of all lawful purposes, which is basically trusting the government to use the technology lawfully. And then control the use of OpenAI actually through the deployment. So there's a few things that they've done, which is it has to be deployed through the cloud and not at the edge. So OpenAI can monitor how it's being used and can actually enforce red lines.

17:38Akash Pasricha:They've got safety and guardrails that OpenAI controls, which is basically OpenAI can enforce the guardrails directly through the model itself. Third, they have their safety and alignment team deeply embedded inside the deployment of OpenAI's product inside the Department of War. So it's really this matter of, you know, Anthropic trying to restrict and get these bright red lines, which there are two things that they really want, which is, you know, no mass surveillance of U.S. citizens. And the second one is no autonomous use of AI in the use of weapons. So those are the things that Anthropic wanted up front and the Department of War committing that they don't do those two things.

18:25Akash Pasricha:But if you think about war and government, it's always a nuanced thing. There are gray lines that humans need to make the judgments. And the government's basically saying, in the Department of War, saying you need to trust that we are going to do the right things for U.S. citizens and our country. And Sam has ultimately decided that he trusts the U.S. government and the Department of War to do that. I would add that OpenAI also added one more red line that is very important to them. And that is no high stakes automated decisions. Things like social credit, where, for example, the CCP and Chinese government monitor whether you jaywalk across the street and you take a hit on your social credit.

19:16Akash Pasricha:And that has implications to your ability to borrow money and perform business activities. is OpenAI has also added that the US government cannot do that. But I think ultimately, I do believe, and I admire Dario and Anthropic for what they've done, which is take a stand for the values that they actually incorporated the company on. And if you remember, Dario and his team worked at OpenAI, and they were very critical in the launch of ChatGPT and GPT-3. And they spun out of OpenAI because they felt that at the time, OpenAI wasn't prioritizing the values of safety, which they cared about. And they've built the company upon the principles of safety.

20:00Akash Pasricha:And so there was a real test. And I was really, really curious to see if Dario actually stood up to the constitution that basically he built the company upon and massive respect to him and Anthropic for doing that. But I don't think ultimately it's a moral argument because I think OpenAI is going to enable the things that the government needs to do, but have the right type of restrictions in place. Right.

20:29Aaron Tilley:I want to zoom out a little bit. So this was not the only news that OpenAI made last week. It obviously had the big funding round. And one thing that we were all sort of studying closely was some of the conditions upon which some of the backers, namely Amazon, have underneath the$50 billion investment. and there's some conditions. You either have to go public or you have to basically reach AGI, I think is sort of the understanding here. And I just want to ask you about the IPO piece of all this. I mean, what is your understanding here of the timeline for OpenAI on an IPO?