Who Really Controls the Future of AI?

There's been a whole slew of labs releasing new models recently. So OpenAI, Anthropic, Gemini, Meta. Is this a coincidence? These companies, they want to upstage one another. They want to try to steal one another's thunder. Hello from New York. I'm Ellie Austin and this is Fortune Daily. OpenAI's Astra model is already creating waves in the cyber security community, as we discussed yesterday, and also in the greater AI landscape. But questions remain on who gets access to this powerful new model and how, as these models increase in capacity, should they all be regulated? Silicon Valley is split over whether legal guardrails around AI would make America safer or slower. So what does a smart AI rulebook actually look like? And can the US protect people without slowing competition? Fortune AI editor Jeremy Kahn joins us now to break down the ins and outs of Astra and the fight over AI regulation. Jeremy, hello. Thank you so much for joining us. So a big day for you today as our AI editor. OpenAI has just released its new Astra model, which we've already previewed a bit on the show and also talked about the fact that this is viewed as being one of, if not the most powerful model out there today. Tell us a bit about the technique used to create Astra. Yeah, so there's been a little bit of a controversy about exactly how OpenAI created this model that sprung up this week. There was a report that they had used a technique called recurrent depth as for part of this model. And basically the concern there is this recurrent depth technique may in the future make it harder for experts to look and interpret what the model is thinking, essentially. There's something called chain of thought monitoring that's used right now to help control AI agents where essentially they output their reasoning in natural language. So if you're asking the model a question in English, it answers in English and tells you kind of what it's thinking. Well, this technique called recurrent depth, if it's extended and if people continue to use this technique, could eventually mean that you don't get that chain of thought. So you can't really see what the model is thinking. You would just get an answer, but not really see the intermediate reasoning steps. And that could be dangerous, AI safety experts say, because one of the main ways we try to tell what these models are doing and what their intentions are is from that chain of thought. And if you don't have that, it becomes a lot harder to know if the model is up to something you don't want it to do and potentially stop it. So that's the concern here. Now, OpenAI has said, look, they've used this technique, but not to the extent that you would lose the chain of thought. The technique is used because it makes the models much more efficient to run, but security experts and safety experts think this is a bad precedent, that it could create a situation where other companies want to extend this technique and then the pressure is going to be on all of these AI companies to use it, in part because it lowers the cost of serving the models. And that would be bad because eventually we'll get to a point where we just cannot monitor what the models are thinking. That's really interesting. Organizations have to qualify, as I understand it, to be in OpenAI's trusted daybreak program, which means that you get access to these incredibly powerful models. How high is the threshold? Hold, who gets in to daybreak? Yeah, that's a good question. OpenAI hasn't been very specific about who qualifies, but it's mostly, you know, sort of Fortune 500 companies, larger enterprises, governments, that OpenAI vets to make sure that they are legitimate companies that want to use this Astra model for primarily cybersecurity, for cyber defense. The model has very high levels of capability when it comes to cyber tasks. The concern is that somebody might use this for hacking or might try to use the model for nefarious purposes. So OpenAI is restricting its release at first to people who are in this daybreak program, which again, they vet all of these people to make sure they're trusted and make sure that they are unlikely to use the model for some nefarious purpose. There's been a whole slew of labs releasing new models recently. So OpenAI and Astra, as we've just talked about, also Anthropic, Gemini, Meta. Is this a coincidence? Was this all times that these new models would be available at around the same time? What's your reporting told you? Yeah, it's a good question. This tends to happen and has been happening, you know, frequently over the last few years where all of the companies release models around the same time. And I do think there's some intentionality to it that these companies, they want to upstage one another. They want to try to steal one another's thunder. And I also think that the industry as a whole, people talk to each other. They generally know when these models are going to be released. Maybe they know the week. They may not know the exact day. So they do try to time these announcements to get out ahead of one another. So you do get this sort of pile up of models being released in the same week. Now, the models do have to be ready. So, you know, I'm sure the labs and the companies are all racing against these deadlines. And they have to actually hit those deadlines in order to release the model. But yes, we saw a slew of models released this week. It's been really quite something. Is this your reality going forward, Jeremy, as an AI editor that every couple of months you'll be bombarded with this collection of new releases? I mean, what do you envisage the future looking like when it comes to a release of models from these massive juggernaut companies? Yeah, it's a good question. I am a little worried that someone has to cover this about how fast things are moving. The companies themselves are a little worried about how fast things are moving. And you've seen this call, at least when it comes to these most powerful models, this call from OpenAI and Anthropic to try to slow down. But both of the companies have sort of said they would like to see some kind of industry-wide mechanism for doing this, perhaps with government approval and maybe a government kind of sponsored mechanism to actually slow down the pace of model development. I think they fear that there may be some antitrust issues if they just try to talk to each other without government approval. But beyond that, I think there's just so much competition among them that they feel unless some outside entity steps in and compels them to slow down, it's very hard to see how sort of market forces make them slow down in any way. And in fact, things are just getting faster and faster. One of the interesting things that happened this week with the model that Google released, Gemini Flash 3.8, this is actually the third of these Gemini Flash models. It's released in just six weeks. And the reason it's been able to do that, it says, is that it's increasingly using AI models to design the next generation of AI models. This is something called recursive self-improvement or RSI. So you have models designing the next generation of models. And because humans are not as involved in the process, it just starts to accelerate. It feels in a way like we've been talking about the possibility of regulation here in the US for a very long time and nothing has yet materialized. It's obviously a different situation in Europe. There are degrees of regulation now in China as well. What do you think the likelihood is of the government stepping in at some point to help regulate these companies? I think at some point it will happen. The question is sort of when will it happen? There were reports going back a couple of weeks that the Trump White House actually was going to come forward with a proposal for a new kind of regulatory agency. Now it was going to be a self-regulatory body, perhaps modeled on FINRA, which is the financial regulator for broker-dealers. They were going to use this as a model. This was an idea that actually Demis Hassabis, the former CEO, now chairman of Google DeepMind, had proposed. And there was lots of reporting that the government had actually taken on this idea and was going to pursue it. Then in the last week, there were reports that actually David Sachs, who is the former AI and crypto czar, he no longer has that role. He doesn't have an official advisory role, but is still very close to the president. There were reports that he had gotten involved to try to kill off this idea. He's opposed to having any kind of regulatory agency, new regulatory agency in the US. He says it's like having a DMV for AI is his line on this. And he thinks it's just going to slow down development, it'll put the US behind in this sort of perceived race with China for AI dominance. So he is opposed to a new regulatory agency and he seems to have managed to gotten the president's ear and won the day for now. But I do think just there's so much concern about, there've been a number, and I know we've spent a lot of time talking about it on this program, these rogue AI incidents have started to occur that are really quite worrying. People are legitimately worried about the cybersecurity implications of these more powerful models. So I do think the balance of momentum is towards some kind of regulation actually happening in the US, but it's just a question of when and how long I think the David Sachs's of the world can fight a successful kind of rear guard action. Jeremy, we'll keep speaking to you to see what happens over the coming months. Thank you so much for joining us. Oh, thanks so much for having me on Ellie. Gloria Steinem has died at the age of 92. Since the 1960s Steinem, a trailblazing journalist and activist, challenged gender stereotypes and campaigned loudly for equality and human rights across all areas of life. In doing so, she became the very personification of the women's liberation movement and inspired younger generations of women to support each other and to fight for their rights to live and work as they choose. In April of this year, I was incredibly lucky to spend some time with Gloria at her gorgeous Bohemian Manhattan townhouse because for the past couple of years, she hosted talking circles in her living room for groups of women. I was there with executives from the Fortune Most Powerful Women community to discuss how AI, of course, is impacting what it means to be an effective human leader. Gloria's main message to us was, as technology continues to evolve at the speed of light, we need to prioritize time together in person more than ever. She certainly practiced what she preached. She clearly loved being surrounded by younger women and she held our hands as she asked us about ourselves. She didn't want us to leave. She let us roam all over her house. And after the talking circle ended, I asked her why on earth she kept inviting groups of strangers into her home in her 90s to discuss the issues of the day when she could be, I don't know, reading a good book or taking a nap. And she told me, I have this house, what else would I do with it? I can think of many other things you could do with it, but Gloria was generous and curious until the very end and she always wanted to be surrounded by groups of women. What an icon and what a pioneer. Gloria Steinem, thank you for everything. Next up, editorial director for Europe and the Middle East, Kamal Ahmed, gives us a preview of what to expect at next week's United Nations General Assembly here in New York. Kamal, hello. Thank you for joining us from London for the day. Honored to have you here. So I want to look ahead to the UN General Assembly, which kicks off in New York on September the 8th. Lots of chatter about what might happen with Andy Burnham, the UK's new prime minister, and President Trump. How likely do you think it is that they will have their first meeting? Well, I think from the point of view of the United Kingdom, they really would love a meeting. And protocol would usually suggest that the president of the United States meets the new prime minister of the United Kingdom. But obviously Donald Trump has changed protocols. He can throw the cards in the air and decide maybe he doesn't want to meet the new UK prime minister. He had a reasonable relationship with the former prime minister, Keir Starmer. But Andy Burnham has got a pretty small international footprint. He's not known for his foreign policies. He's not known for his sort of global arguments. And Donald Trump might think under the United Nations General Assembly is not a moment for that. But for the United Kingdom, it's a real signal of, is Andy Burnham a serious player on the world stage? I think the biggest weakness for him as the new United Kingdom prime minister is that he doesn't have that foreign policy hinterland. I remember many, many years ago, a former leader of the Conservative Party, the main right-wing opposition in the United Kingdom until the rise of Nigel Farage. And he said to me a really interesting thing, that the British public, the British voters, don't take a leader of a political party seriously until they can imagine him or her walking across the White House lawn and shaking the president's hand. Andy Burnham needs that moment. Will he get it? I would say 50-50. But do the UK want it? 100%. What sort of background machinations are going on now on the UK side to try and make this happen? I would imagine that they are trying to show that Andy Burnham is not coming for a kind of gunfight. He's not the Mark Carney figure of Europe. He's not going to be obdurate about how he approaches the president of the United States. There's far too much downside for the United Kingdom in a battle with the United States over tariffs, over international policy. He's going to have to do the same thing as Sir Keir Starmer did, which was tuck under, sometimes hold your nose. They're not natural bedfellows. Andy Burnham is leader of the Labour Party in the United Kingdom, of course, similar to some parts at least of the Democrats. Donald Trump is the MAGA president. But he will have to show that he's going to be a reliable ally. And they will just be signalling that from what we have heard from the UK, there's no change of direction from him than there was with Keir Starmer. Now, of course, Keir Starmer had the big prize of bringing the state visit and the invitation from King Charles, which Donald Trump really, really liked, played to Donald Trump's ego. Andy Burnham doesn't have that card to play. And also the president might consider that UNGA is not the right place to do that first meeting. There are lots of global leaders in New York for that. And he might say, we won't do it at that moment, but we'll have it as a specific moment at the White House, which is, of course, where he met Keir Starmer. Andy Burnham really reminds me of Mamdani in his approach to social media. He's the everyman. He's Northern. And he's really playing into his Northern personality. He's out there. He's pulling pints. What's the reaction to him? What's the temperature right now in the UK? He's a few weeks into his term at this point. I think it's pretty warm. I think that he has come in. He was such a contrast. He has been such a contrast with Keir Starmer, who was seen as stiff. He's a former lawyer. He seemed to sort of put legal ahead of politics. He didn't seem very warm. Andy Burnham, as you say, is the complete opposite. He's very live on TikTok. He, as you say, he goes to local pubs. He was seen at the football, the first weekend of the Premier League. Recently, very clearly, not in the director's box where Keir Starmer used to stand to watch his beloved Arsenal. Andy Burnham follows a team called Everton in Liverpool. He was in, of course, the expensive seats, but he was in the stands. Very different type of person. And I think the response has been few. Someone that we can relate to. The other big thing for Andy Burnham, which I think Keir Starmer never put across. I don't know whether it was true or not, but Keir Starmer never put across. Andy Burnham is clearly enjoying the job. He comes to it as if, this is a real privilege. It's a real honour. And I really want to do it. And I think a lot of the public, whether that's in America or in the UK or across Europe, they want to see someone who really wants to do the job. Whether I agree or disagree with what he's saying, he seems to be a person who is engaged and wants to do things that are clearly defined. The other thing he's had, beyond his TikToks and beyond the skills he has as a ordinary bloke, which has some traction, is that he has dealt with everything to do with cost of living. Like Mamdani, real bread and butter issues. The types of things that are spoken about around the kitchen table in the evening. Energy prices, the price of food. Those points have been landed by Andy Burnham. He's done some small things and that means that actually the polls have shown Labour are back ahead for the first time since the general election in 2024. Which is huge. Kamal, fascinating stuff. I'm sure we'll chat again during UNGA. Thank you so much. Now, the future of work and Executive Editorial Director Diane Brady tells us about how humanoid robots are revolutionising logistics firms. Diane, hello. Thank you so much for joining us. I am so thrilled to be in your alley. So Diane, you are the author of Fortune's CEO Daily Newsletter and you recently reported on a logistics company that is using humanoid robots. Educate me, first of all, on exactly what a humanoid robot is. Well, I mean, when you think about it, you can either think Terminator, Blade Runner, Ex Machina. So let's put that aside for a second. Basically, they are robots that mimic human behaviour. So they're bipedal. They can either walk or be on wheels. But the idea is that they will have what's called in essence the range of freedoms that humans have. So it's their ability to walk inside a minefield, to walk inside a factory. That's what we're looking for because the problem with industrial robots is they usually have one function. They're fixed. So we're trying to free up the robots to be like us, which of course creates a host of new opportunities and challenges. It does. And you spoke to the CEO of GXR. Now tell me a bit about why they specifically are integrating robots into the workforce and what these robots are doing. So they are the largest pure play logistics company in the world. They've got about 150,000 people. And logistics is one of the best use cases for robotics and for humanoids in particular, because they have to lift heavy pallets. They have to sort lipstick. Basically, they're trying to do something that humans do in terms of moving around the plant. The other thing is it's also an industry that's very short on talent. People who work in these industries tend that they have high turnover because it's grueling work. So they're doing a lot of experiments and pilots right now with really trying to come up with the perfect humanoid robots. And it turns out there's a lot of challenges, one being the hand. The dexterity. Well, basically, the hand has about, I think, 25 ranges of freedom. I think that's what it's called. And right now, the human body has 200. Most robots, humanoid robots, tend to have around 35 to 45. Some have more than 100. The closer we get to 200, the more the robots will mimic us, Ellie, and the more they'll be able to work. But the cost is coming down. The safety is going up. They're doing five, soon to be six pilots. And in the next year or two, we're going to be seeing probably a lot more humanoids, specifically in logistics, but maybe around the world. And the CEO of GXO told you that the emergence of humanoids in his workforce isn't going to result in actual humans losing their jobs at his company. Do you believe him? Well, I think let's, let's have a moment of pause for that, because I, no, I think ultimately, of course, much like we've seen with factory work, we're going to have humans replaced. But this is an industry where if you have 20% turnover, you're lucky. So what I imagine to be the case is when there's attrition in these, that they will not replace them. I do think it will allow the humans to do more high value work. But, you know, if you look at the economics of humanoids right now, they're expensive. But a lot of industry analysts think it's going to be $10 to $12 an hour to operate these things. Most humans get more money than that. So the economics is such that if you can use humanoids, then I think ultimately, will there be 150,000 people at GXO? Maybe, but it might be a bigger company and it might, they might be deployed in very different ways, different functions. And you mentioned that humanoids are maybe going to move into other sectors, which industries beyond logistics do you think are particularly ripe for the arrival of these, of our little robot friend? Well, I think definitely defense, you know, in the same way that drones kind of can take, you know, bombs in safely to, maybe not for the recipients, but I think one of the things with search and rescue, for example, these very dangerous human tasks, that's one area. Frankly, companions, you know, you think about, there's a shortage of healthcare workers, especially home healthcare aides for the elderly. If you can have humanoids, you need that level of dexterity to make a sandwich, to make a bed, to maybe assist somebody getting out of the bed. So that's ultimately going to be a very strong use case. But one of the challenges as well as safety, and safety being humanoids by their nature are kind of, they're freed from being constricted in certain zones. So when you have robots that are moving around with humans, you need them to have the same sensitivity and fluidity that we have in the same nuance. So it's a little early yet to be freeing up a humanoid to take care of your grandmother because that could end badly. It feels a bit dystopic, I have to say. It is a bit dystopic. I've seen Blade Runner, it's a cool movie. I wouldn't want to be the replicants. But I think ultimately what we're looking at is using essentially the essence of AI and technology just to do the things we don't want to do and to do the things that frankly a lot of humans find dangerous to do. So in the one hand, I'd say it's dystopic. In the other hand, it's going to be a huge growth area and we're seeing already record profits for companies. I would take a close look at the logistics space, the warehousing space. Those are places that are really going to benefit from some of the advances we're seeing right now. Diana, where can we find all of your reporting? You can subscribe to CEO Daily. It comes out every day. And do you know, I try to do as many interviews on stage and off and you do that for the magazine. But CEO Daily really is, you know, it's a great joy to do. It's where I live and it's really about making people smarter about what's around the corner. Thank you so much, Diana. Great to be here. That's it for Fortune Daily. I'm Nellie Austin. Please remember to like, subscribe and to let us know what you thought of today's subjects. We'll see you next time.