A Big Shift in the AI Race

Today on the AI Daily Brief, some big shifts in the AI race. Before that, in the headlines, the latest on Anthropic's fight with Washington. The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Robots and Pencils, Mission Cloud, and OutSystems. To get an ad-free version of the show, go to patreon.com/aidailybrief, or you can subscribe on Apple Podcasts. If you want to learn more about sponsoring the show, send us a note at sponsors@aidailybrief.ai. We are back with a normal episode today. Appreciate your forbearance yesterday as we got to go make some core memories at the World Cup. And of course, the big story happening behind the scenes in AI continues to be the fallout of the Fable 5 shutdown and the negotiations to try to bring it back. It has at this point been 5 days since the US government shut down Mythos/Fable, or rather imposed the conditions upon Anthropic where they were forced to shut down Mythos/Fable, and it doesn't particularly appear like we're any closer to a resolution. On Monday, Anthropic leaders touched down in Washington to explain the situation and to try to come to some new type of understanding. Now, interestingly, and I think giving some insight into how Anthropic is viewing this situation, rightly or wrongly, is that the talks appear to have been largely technical. Anthropic sent Chief Compute Officer Tom Brown, Head of External Affairs Sarah Heck, Head of Red Teaming Logan Graham, and Senior Security Researcher Nicholas Carlini. The Wall Street Journal actually ran quite a long piece about Carlini titled "The Hacker Sent by Anthropic to Calm the Government's Nerves About AI Safety." The journal describes him as a well-respected hacker who is considered something of a professional skeptic of AI cybersecurity claims, who has, quote, "lately changed his mind." Putting some fuel on the fire of no, this time was actually different, they write, just weeks after getting his hands on Mythos, Carlini offered a stark warning to a standing-room-only crowd of cybersecurity experts. First, he showed them how he had used Anthropic's AI to find and exploit a critical bug in a piece of web publishing software called Ghost. Then he demonstrated another in the Linux operating system, one of the most battle-tested pieces of software, which powers billions of devices. Carlini had never before found a bug in Linux or in Ghost. Ghost. But now he had discovered many. What he was seeing represented a new world order for cybersecurity. The balance that existed between attackers and defenders over the past two decades, quote, "seems like it's probably coming to an end." He continued, "It's pretty clear to me that these current models are better vulnerability researchers than I am." Now that was back in March, and it sounds like Carlini was one of the folks inside of Anthropic pushing the company not to release Mythos at that time. Fast forward to the situation we have today, former DOD official and Council on Foreign Relations senior fellow Michael Horowitz said, The government and Anthropic clearly have an inability to communicate effectively with each other. More technical exchanges should be helpful in socializing these issues in a way that should lead to better decisions. So coming back to the negotiations, on the government side, staff from the Center for AI Standards and Innovation in the Office of the National Cyber Director also attended the meetings. Commerce Secretary Howard Lutnick phoned in from the G7 summit in France. However, it appears that none of the key architects of the decision, including National Cyber Director Sean Cairncross, Treasury Secretary Scott Bessant or White House Chief of Staff Suzy Wiles were in attendance. Given the prior reporting, then, it's questionable whether anyone in the room on Monday had authority to actually call off this Mythos Fable ban. And one thing that is worth underscoring at this point, obviously, since so much of the coverage, mine included, has focused on Fable because gosh darn it, you took our new toy away— this is absolutely not just a Fable 5 ban. It is very much a Mythos ban as well. In fact, it seems like the re-release of Fable to consumers is fairly low on the list of the government's concerns. Sources said the Commerce Department expressed a willingness this to allow Fable to come back online as long as Anthropic fixed the jailbreak that was at the heart of the matter in the first place. Mythos, it seems, is another question. Over the weekend, we had some very thin reporting from Semaphore that the administration feared the Chinese government had gained access to Mythos. And on Tuesday, the Washington Post reported that a big part of the schism with Anthropic had come about from the expanded access to Mythos a few weeks earlier. The WP wrote that Anthropic added around 50 firms to Project Glasswing at that time, And it was then that, quote, senior officials began to consider using export controls to claw back the technology after Anthropic did not identify the new recipients for days. Once the government had the list, they discovered that one recipient was a South Korean telecom company that the administration suspects of having ties to the Chinese government. Now, the nature of the jailbreak is also getting increased attention in the press. On Tuesday, The Atlantic elevated the analysis of Ludda Security CEO Katie Moussouris, which I discussed in earlier coverage. They rehashed the point that Fable will refuse to review insecure code for security issues, but when asked to fix the bugs, it will generate patches. Moussouris's point has been that this was just the model working as intended for cyber defense, arguing that GPT-5.5 and Opus 4.8 work in the same way. And yet, of course, as we've seen over and over, part of the meta story is about personal interactions between the major players rather than technical specifics. In Monday reporting, an administration official told Axios that Anthropic knew the model was susceptible to a jailbreak and decided to distribute it anyway. They came to every fork in the road, the source said, and took the wrong fork. A source familiar with the administration's thinking commented, "Anthropic has not done a great job at trying to speak to the administration and appreciate the ideological differences. It's like they just speak in different languages." Another quote attributed to an administration official put it more bluntly. Everybody said Anthropic was a bad actor. Some of us said it was time to give them a chance. Now those people are questioning that. They screwed us. As the week has gone on, we've gotten more and more details. On Tuesday, Bloomberg published the full text of the letter from the Commerce Department that shut down Mythos and Fable on Friday night. It's written largely in legalese, but the thrust of it was that Anthropic risked criminal and civil penalties for non-compliance. The order required them to remove access to all foreign nationals wherever located. And when during his Friday night phone call with Commerce Secretary Howard Lutnick, Dario Ahmed reportedly said, "This means we can't have the model out," Lutnick responded, "That's the point." Charlie Bullock of the Institute for Law and AI commented, "The legal theory strikes me as strange, very aggressive, and probably vulnerable to legal and constitutional challenge." That said, I don't think that Anthropic will actually sue to challenge anything. Instead, I think they'll work to resolve their differences with the government without resorting to litigation, and that the government will eventually allow them to publish a version of Fable again. Now, Charlie also got into one of the bigger issues here, which is the terrible implications of having a regulatory regime that's not really a regulatory regime. Bullock writes, The takeaway here, as I've said before and will say again, is that this ad hoc, last-minute licensing regime is bad for companies and for the Trump administration and for the public, and that we desperately need to replace it with legislation. It's no good for the administration's stated goals that their actions are vulnerable to legal challenge. They should want to fix that problem as much as Anthropic does. Now, I actually think that Ashley on X, @Polyletheia, gets it directionally correct when she writes, I think I see what's going on here. Anthropic is negotiating with a regulator without realizing it. When you submit something risky to a regulator, you have to, one, describe what the risk is, and two, show how you're going to mitigate that risk. Importantly, you have to properly estimate the scope of the risk. If you scope the risk too broadly, then you place extra burden on your mitigation. If you scope too narrowly, then you run the risk directly. But you must commit to an estimation to the regulator. What happened here is that Anthropic scoped too broadly, and when it was shown that their mitigation was insufficient for the stated risk scope, they responded by trying to narrow it. This is a huge red flag to any regulator for good reasons, so regulatory action followed. And here's what this kind of comes down to for me. I think it's pretty undeniable that the government is telling on itself left and right, that it doesn't really have the technical capacity, at least with who's in these rooms right now, to understand anything that's happening. We could be mad at that all that we want. You could also be mad about who's in power. In fact, being mad at that is one of the privileges of democracy. What you don't get to do as a company that is now seen as integral to the economy and national security, and thus a major political consideration, is fail to recognize that a big part of your job, a significant part, in other words, of what it means to run the company, is to deal with the government in power. I think the most charitable interpretations of Anthropic's actions is that they just haven't realized the extent to which this was now their job. Look, there are lots of little details that seem at least a little bit like political gamesmanship, such as the planted detail about Dario being at a wellness retreat when they were trying to get in touch with him. But there is also a pattern of Anthropic not responding particularly quickly to various concerns and contacts. There are reports that Amazon's Andy Jassy reached out to him before he reached out to the US government. We've heard that when they expanded Project Glasswing a few weeks ago, it took them days to get the list of the new actors to the US government. And when even the counter to the wellness retreat narrative was that it took Dario an hour and 15 minutes to get back to the US government on Friday as this was all going down, if that's true, what the hell were you doing for the other hour and 15 minutes? If the government calls with something that's clearly this mission critical, that's not finish the meeting that you're having. That's pause whatever you're doing and get on the phone. So again, I think the charitable interpretation is that Anthropic just hasn't realized up till now that managing their relationship with the US government is now as important a job as anything that they build in the lab. So where does this leave us? Cybersecurity experts continue to believe that this is a dangerous misstep from the administration. And have said in an open letter with more than 100 signatures now that removing Mythos from the cyber defense toolkit makes everyone more vulnerable. Some notable geopolitical analysts believe this is a gift to China. Agatha Demarais of the Council on Foreign Relations wrote in the Financial Times that, quote, "Washington has just done more to boost the appeal of Chinese AI models than Beijing ever could have hoped." The business world is concerned about the precedent. Jim Reid, the global head of macro at Deutsche Bank, wrote in a Monday note that if the ban continues, quote,— it's not great news for US tech firms or for those assuming breakneck speeds of AI adoption. You can't rely on something that could be switched off. And finally, the ban has sent the US-based AI research community into a tailspin. Citing a source close to OpenAI, the Financial Times writes that in recent days the industry has been working on ensuring foreign national researchers could continue to work on developing the most advanced models, a practice, as they put it, that the Anthropic directive has now banned. Now, despite the multidisciplinary outcry, at this stage it seems the administration is sticking to their guns and demanding a patch for the jailbreak. For those familiar with the technology, it's not even clear that that's possible. Helen Toner, a former member of the OpenAI board, said, It's a pretty widely agreed upon fact that you cannot fully fix jailbreaks in these models. It's a very inexact science. Now, while at this stage it seems like negotiations in Washington have kind of stalled out, this morning CNBC reported that AI leaders including Dario Amadei had traveled to France to take part in the international G7 summit this week. And Politico even reports, that according to the official agenda, there will be a 2.5-hour lunch with CEOs including Dario Amodei, Sam Altman, Demis Hassabis, and Mistral's Arthur Mench. And while the official agenda is all about economic growth and resilient societies, it's hard to believe that the most important discussion won't be what's happening in Washington right now. We'll see if anything interesting comes from that, but for now, that will close our Anthropic headline report today. And next in the main, we'll talk about how Elon Musk is taking advantage of the moment to make some serious moves. One of the most important AI questions right now isn't who's using AI, it's who's using it well. KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising. The highest impact users aren't better prompt engineers, they treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com/us/sophisticated. That's kpmg.com/us/sophisticated. One thing I keep seeing in enterprise AI: companies hedging across every cloud, every model, every framework. Or paying a GSI for a pilot that never ends. The team's actually shipping, they've picked a lane, and they move fast. That's one of the reasons I like today's sponsor, Robots and Pencils. They've gone all in on AWS. They're an advanced tier and AWS pattern partner, and they ship production AI coworkers in 45 days. That's led to them doing some of the more interesting work I've seen on AI coworkers. And by that, I'm not talking about chatbots. I'm talking about actual agentic systems that sit inside a business architecture and do real work. That kind of focus matters if you're an enterprise leader trying to get something real into production or an AWS rep trying to move a customer from interested to deployed. Request an AI briefing at robotsandpencils.com. One conversation with Robots and Pencils and you'll know. The average enterprise is spending $11.5 million on AI this year, and most of them can't prove a single dollar came back. What does AI actually look like when it produces ROI? Ask the healthcare company that just made their payment processing 320 times faster, or the law firm whose document research went from 3 months to 10 minutes, or the contact center who reduced wait times by 99%. These are real Mission Cloud customers with real results. Mission Cloud is a CDW company and an AWS Premier Tier partner. They're the AI-first, outcomes-obsessed AWS experts who build AI solutions that drive your business forward. Whether you're flooded with AI ambitions but no idea where to start, or 6 months into a deployment that's going sideways, they've seen it and they've fixed it. Stop burning your budgets on AI that doesn't produce results. Start at missioncloud.com. This episode of the AI Daily Brief is brought to you by OutSystems, a leading agentic systems platform built for the enterprise. Organizations all over the world are building, orchestrating, and governing agentic systems on the OutSystems platform, and with good reason. OutSystems' open and unified platform allows teams to architect, deliver, and scale governed agentic systems with agility. Teams of any size and technical depth can use OutSystems to build, deploy, and manage AI apps and agents quickly and cost-effectively without compromising reliability and security. With OutSystems, you can rapidly launch ideas from concept to completion. It's the leading agentic systems platform that is unified, agile, and enterprise-proven, allowing you to accelerate growth, reduce operational friction, and deliver real enterprise impact with AI. OutSystems. Build your agentic future. Welcome back to the AI Daily Brief. Today we're talking about something of a shift in the AI race. And while this isn't directly related to the Fable 5 controversy and Anthropic's dealings with Washington, it certainly does feel like we are in a moment of realignment. Yes, there is of course a political dimension of that, as we are all realizing, but the realignment is also happening as the economics of mature agentic AI start to become clear. As I continue to discuss, companies are starting to grapple with the reality of what full agentic workloads are going to cost, and they're quickly working to figure out how to become more efficient with their token consumption, whether it's through lower-cost models, better routing, or in the case of more poorly considered strategies, limiting who has access to certain types of advanced AI. Meanwhile, one of the downstream implications of the economics of AI labs is the broader state of the US economy. A huge part of GDP growth is related to the AI infrastructure buildout, which is of course justified by increases in lab revenue. And TL;DR, all of these things are happening at the same time. We are in a very liminal moment. Now SpaceX specifically and Elon Musk more generally have had an interesting place in all of this. On the political side, he was cozy with this administration to start, although that inevitably ran into conflicts fairly quickly. And when it comes to AI, his strategy for a time didn't particularly look like it was working. While Groq has been beloved by some, it is generally not considered competitive with the top models from OpenAI and Anthropic. The folding of xAI into SpaceX at first seemed to at least a fair number of market commentators as something of an xAI bailout, but then everything shifted. It turns out that when you build two huge supercomputer data centers like Colossus 1 and Colossus 2, that's an asset that you can monetize fairly easily in this environment. Over the last couple of months, Elon and SpaceX signed deals to give Anthropic and then Google access to that compute, And all of a sudden, the economics of SpaceX started to look very different. Basically overnight, Neo Cloud revenue became their number one source of revenue, and as they looked to space for the next generation of orbital data centers, the narrative heading into their IPO kind of made a lot more sense. Now, whether you think it was that sort of considered analysis or just SpaceX being the biggest meme stock of all time, one thing that is undeniable is its performance post-IPO. On Tuesday, the stock closed at $201.80, up —49% from the IPO price. The company is now valued at $2.6 trillion, making it the 5th largest company in the world by market cap, slightly ahead of Amazon. Now, for many observers, they just can't get over Elon Musk's personal wealth. On the back of the IPO, he became the world's first trillionaire, and he's now on the order of 3 times ahead of the next closest of the world's richest people. Now, of course, most of his wealth is tied up in his 46% stake in SpaceX, And even without those lockups, he couldn't really liquidate any sizable part of that stake without absolutely tanking the market price. But still, the sheer numbers here are so gigantic that a lot of people just can't think about anything else. Flexport CEO Ryan Peterson pointed out on Monday, "With today's 20% SpaceX pop, Elon made more money today than Warren Buffett made in his entire career." Ryan then came back to clarify, "I posted this before realizing it was up another 14% after hours." Now the SpaceX IPO is going so well that some are suggesting that they might have even left money on the table and could have raised even more. For others, the whole thing is a farce. Twitter is absolutely awash with Big Short video clips that having SpaceX's valuation ahead of Amazon when Amazon has about 40 times the revenue of SpaceX just seems nuts. Others point out that the vast majority of SpaceX stock is currently locked and that as it becomes unlocked over the course of the rest of the year, that could create its own sort of bearish pressure. But for now, none of that matters. And it's clear that when it comes to the AI race, Elon is going to take advantage of SpaceX's new fortunes and the opportunities it affords him. Investor Bill Ackman wrote, "One of the things that makes SpaceX so valuable is how valuable it is. The Cursor acquisition," which of course we're about to talk about, "costs materially less in dilution because of SpaceX's high valuation. SpaceX's ability to do economically, strategically, and technologically accretive acquisitions is an important component of its value. There is enormous value inherent to a company with a high value, particularly when it is controlled by an entrepreneur that the most talented people want to work for and partner with. Value begets value, talent begets talent. Now this discussion was around the culmination of the Cursor acquisition, which SpaceX had announced the right to engage in earlier in the year. The deal did come in at $60 billion and Cursor will become a wholly owned subsidiary of SpaceX. As part of the announcement, we found out that Cursor had hit a $4 billion run rate and was growing 7x year over year. But what's much more interesting is what Cursor's going to mean for SpaceX's AI strategy going forward. As I mentioned at the beginning, the shift of focus on their Neo Cloud strategy has clearly become a more important strategic priority for SpaceX over the past few months. But the Cursor acquisition potentially changes a bit about where SpaceX, or rather xAI, might fit in the model and enterprise AI battle. Now going back to this transitional moment between token subsidy and token scarcity, Cursor identified at the end of last year that they could not just play the harness game, they needed to play the model game as well. Since then, they've released a series of models under their Composer brand with their most recent Composer 2.5 performing in the same range as Opus 4.7 and GPT-5.5 at about a tenth of the cost. Now that model was largely about post-training using a KIMI-K model base, but now Cursor is teasing a new trained-from-scratch model. At the Compile event, engineer Nick Dobos wrote, new cursor model being teased at Compile. Same size as Claude Opus and GPT-5.5. Trained from scratch. No more Kimi Base. 10 to 20x more compute versus Composer. Generally intelligent, not just coding. Releases in the next couple of weeks. Some people are extremely bullish about the possibilities. Lassan on X writes, I expect SpaceX AI to be between Google and OpenAI by the end of the year. Composer 2.0 was a very strong model, but Elon isn't stopping at 1 trillion parameter models. The Financial Times suggested that buying Cursor could be SpaceX's Instagram moment, referring to the idea that one of the smartest things that Mark Zuckerberg ever did was instead of trying to compete with the hot new thing, just took it off the market very early before it could become a competitive juggernaut. Now in my mind, what'll be interesting to see is whether this new model continues to try to prioritize efficiency in the way that the previous composer models have, or whether this is just xAI and SpaceX getting back into the game, trying to come back out on top and compete at the state of the art, whatever the cost. Investor Chamath Palihapitiya thinks that despite Cursor's recent focus on models, it is in fact their harness and their relationship with customers that matters. He tweeted, This is the first, but not the last, big exit at the application layer of AI. As product value accrues and accelerates upwards, the focus over the next few years will be firmly on the control plane. What gives organizations who want to go all in on AI the governance, control, auditability, and business continuity across models and across time that they will need to firmly make the leap. This is the next big phase of AI value creation that the SpaceX-Cursor merger is highlighting. Now, one more small note on xAI, especially given that the lingering background of all of our discussions right now are the goings-on in Washington. The US government actually recently intervened in a lawsuit against xAI, claiming that Grok is vital to national security. The lawsuit was filed by the NAACP in April. It argues that xAI is in breach of the Clean Air Act by operating unpermitted gas turbines at their Colossus 2 data center. On Wednesday, the DOJ joined XAI's motion to dismiss the case. Their filing stated that the lawsuit, quote, "threatens American national, economic, and energy security by seeking to shut off the power supply for artificial intelligence innovation that supports the Department of War's military operations." A supporting filing from a Pentagon official explains that Grok was used to, quote, "support vital national security missions." This includes apparently targeted decisions for recent strikes on Iran. Now, of course, Grok isn't the only model being used to support warfighting. We know that Anthropic, Google, and OpenAI models are also cleared for classified use. Even Dario Amadei recently confirmed that Claude was being used for missile targeting, but said that he didn't know the specifics. Chubby on X wrote, In one June week, two things happened that look unrelated but are actually the same story. Commerce Department forced Anthropic to disable Fable 5 and Mythos 5 for every foreign national, while the DOJ went to court in Mississippi to defend XAI's unpermitted gas turbines as too vital to national security to shut off. Why is this interesting? Because it clearly demonstrates one thing: AI and everything that goes with it—data center expansion, frontier models, access, etc.—is increasingly being placed under national security and control. Now, bringing it back to the Model Lab competition itself and potential changes therein, Anthropic is of course not the only company with big, with big financial plans this year. Both Anthropic and OpenAI have filed confidentially for an IPO later this year. And of course, in anticipation of that, we expected to start to get a lot more insight into the company's financials. Well, it happened a little bit before OpenAI would have wanted, with inveterate AI skeptic Ed Zitron publishing OpenAI's fully audited numbers for the past 2 years. For 2004, the financials stated that OpenAI had $3.7 billion in revenue, $12.4 billion in costs, and a net loss of a little more than $5 billion. In 2025, OpenAI had around $13 billion in revenue, just shy of $21 billion in operational losses, and a net loss of $38.5 billion. Now, to the skeptics' view, those $38.5 billion in losses were, quote, "astronomical" and far higher than most believed it would be. And yet, when the numbers were shared with the Financial Times, they came to a very different conclusion. While, yes, acknowledging the view that costs were outpacing revenues, they highlighted comments from OpenAI that the majority of the 2025 costs came from a $30 billion, quote, "non-cash accounting change linked to the company's previous structure." In other words, the losses were an accounting entry to reflect the conversion to a public benefit company. Under US accounting rules, revenue sharing rights under the old structure are treated as a liability and are not expected to continue moving forward. According to OpenAI, when you strip away this one-time accounting charge and stock comp, OpenAI lost $8 billion in 2025, which is not nothing, but certainly a very different story. And frankly, when people dug in, the number that seemed to be most overlooked by both of these reports was that OpenAI is apparently turning a tidy profit on inference. In 2024, they generated $3.7 billion in revenue on $2.7 billion in cost of revenue, In 2025, they generated $13 billion in revenue on $7.5 billion in direct costs. Now, obviously, it is too simplistic to strip away training, marketing, and staffing, but for those who aren't just trying to find reasons to be skeptical, that's a pretty promising sign that there's solid profit margins in the core business of selling tokens. It also looks like OpenAI is holding their burn rate steady as the business expands. The information reports that OpenAI spent $3.7 billion in the first quarter of this year, which is a slightly faster annualized rate than 2025, but not by much. Now, that burn rate does not include training costs, which amounted to $8.3 $6 billion in Q1, but the company currently has $73 billion in cash and marketable securities on their balance sheet, which is up from $40 billion in December. As we try to figure out what happens next, this could change some amount of the calculus around IPO. Sam Altman has gone out of his way to say that they have not committed to a timeline for going public yet, and it's not hard to concoct a scenario where between SpaceX unlocks and a dip in the narrative that comes with a dip in the SpaceX price, plus continue weird alignment issues with the US government, that OpenAI continues to take advantage of its cash position and stay private a bit longer. So the point of all of this is that just as it seemed like we were settling into this new agentic phase, it's actually quite clear that the AI race such as it is is an incredibly dynamic and fast-moving environment. The big obvious things to watch for next are: 1, the resolution of Anthropic's Fable-5 issue; 2, whether OpenAI can avoid its own issue like that and actually get out a more advanced model; 3, what SpaceX's public price does in the next few months, and 4, what the next model to come out of Cursor says about what SpaceX's strategy might be going forward. Then of course there's always Google who has been fairly conspicuously quiet this year, and you can't imagine that that's gonna stay the case for long. For now though, that is gonna do it for today's AI Daily Brief. Appreciate you listening or watching as always, and until next time, peace. Peace.