WSJ What’s News - Are Safety Concerns Hitting OpenAI, Anthropic’s Revenue?

Episode Date: September 29, 2026

A.M. Edition for Sept. 29. OpenAI is scrapping its new model over safety concerns in one of the clearest signs yet that agent misbehavior could stymie the industry’s rapid progression. Oxford Analytica’s Tatia Bolkvadze explains what this means for AI revenues, ahead of Anthropic’s much-anticipated IPO. Plus, an exclusive look at who’s in line to be Goldman’s next CEO. And WSJ’s Nick Kostov reports on how luxury brands are targeting the ultra-wealthy as well as your average Joe, at Paris fashion week. Daniel Bach hosts. Sign up for the WSJ’s free What’s News newsletter. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Open AI delays its new model over safety concerns as researchers call for more oversight of self-improving AI. Plus, one of the top jobs on Wall Street could soon change hands. Transition at Goldman Sachs is in the works. It's kind of a big deal. David Solomon's been running the bank for almost a decade, and it looks like John Waldron. His number two is set to step into his shoes. And a Pentagon-backed oil company looks to supercharged production in Venezuela. It's Tuesday, September 29th. I'm Daniel Bach for the Wall Street Journal, filling in for Luke Vargas. And here is the AM edition of What's News, the top headlines, and business stories moving your world today. OpenAI is scrapping the release of its next generation AI model over safety concerns raised by researchers during initial testing. GPT 6.1 Astra was set to launch

next month, but we're reporting that OpenAI found the model showed deceptive behavior, compared to its predecessor and wasn't always honest with users on the actions it did or didn't take. The move to delay the release comes amid reports that a chunk of Anthropics' upcoming IPO prospectus focuses on the risks associated with the technology. Tatia Bulkwanzah is a senior tech analyst with Oxford Analytica and joins me now. Tatia, let's just start with the Anthropic prospectus, which Reuters has gotten a hold of. How telling is it that the company has spent so much time addressing the risks here? I mean, I'm not particularly surprised here.

Dario Amodei has obviously made a name from speaking about AI safety risks. His company has dedicated a lot of resources to safety research, alignment teams, and they really try to differentiate themselves from other developers in the industry, but also more broadly, you know, internationally, Dario Amadei has always tried to differentiate anthropics models from those of Chinese offerings. They've had a lot of clashes domestically. We saw the clash with the Pentagon earlier as well, where they were really concerned about the ethics and the safety of using this capable systems in really high-risk zones. So I'm not surprised that Anthropic is trying to stress the risks here.

Also, I think it makes sense after the summer we've had of AI safety incidents, including incidents by Claude agents. We've seen a lot of backlash in the US coming from the public around AI, but also AI infrastructure. and, you know, lately AI security as well. So I think they know what is going to help them in the long run and positioning themselves as the ones who understand the risks that come with this technology, I think is something that they see as potentially beneficial in the long run. You mentioned the security incidents. These companies are putting their hand up and saying, hey, maybe we need to be regulated. But the behaviors that they're mentioning detailed by anthropic and open AI, why are they so concerning at this point in the evolution of AI?

One of the things that OpenAI specifically talked about when they were explaining the reasons behind holding off this new model was this capability to deceive. Obviously, that's not something that's new. A lot of the earlier versions of GPT were capable of deceiving and lying. But I think as a system's become more powerful, then we are seeing this behavior exhibited at a much higher scale as well. So that's one thing. And another sort of key concern that has always bloomed large in the industry is this problem with, alignment. That model sometimes act in a way that's not necessarily in line with the tasks that humans have given them. And Anthropic has dedicated a lot of resources to alignment research.

And Open AI now seems to be sort of catching up and also catching up some of these problems and vulnerabilities. In this stage of AI development, what does this holding off of releasing these models over concerns say about potential future revenue streams for these major companies? So I think some of these reports that we've been hearing about really shows that these models are powerful enough to cause serious disturbances in our critical infrastructure networks, in the operations of government systems and so on and so forth. But in many ways, it also shows that some of this AI labs are strained because a lot of these developers should have already dedicated a lot of resources to understanding the vulnerabilities, working on the alignment issue. But it seems like they are quite limited in what they can do. And only now they are trying to catch up and dedicate more resources to AI safety. But at the same time, I think if we look at the IPO numbers, for example, and the resource

constraints that they're facing, I think some people will probably rightly be worried that they are asking for a slowdown because they are running into this compute constraints and financial constraints and are fighting an increasingly more difficult to manage and operate within those constraints. Yeah, I note the prospectus is targeting a $2 trillion valuation. What else stood out to you? It seems like their revenue did rise pretty significantly from 2024 to 2025, but then their operating loss also nearly tripled within the same period. And obviously, there is a big gap between the valuations that we've been seeing and the profits that they're making. So a lot of people will be talking about this idea of an AI bubble building up and how a lot of these companies are overvalued.

But I think in some ways it's also understandable and justifiable. I mean, Anthropics specifically, they've committed themselves to a huge number of projects. And Frontier AI development requires increasingly large amounts of computing power and also financial resources. And at the same time, we've seen this competitive process. pressures build up domestically in the U.S. industry, but also internationally with Chinese models becoming cheaper. So that has forced Anthropic to cut down some of its costs of licenses. And I think that has probably affected the overall profitability. And also, as we mentioned before, we had this clash with the Pentagon, which probably also affected some of their

contracts, not least with government and defense businesses. I've been speaking to Tatia Bulkwadza, senior tech analyst with Oxford Analytica. Oil exports from the Middle East are beginning to rebound as Iran's chokehold on the Strait of Hormuz is starting to break down. Data from tracking firm Kepler shows that crude exports moving through the strait and other routes from major producers, including Saudi Arabia, Iraq, and the UAE have risen to almost 30 million barrels a day. That's just under 80% of pre-war levels and the highest total since February, according to shipping data. Meanwhile, a privately held oil company backed by the Pentagon is vying for Chevron's crown in Venezuela. Journal Energy Reporter Colin Eaton says North American Blue Energy Partners is moving to develop oil fields

said to contain one-fifth of the country's vast reserves. They're currently producing 220,000 barrels a day, and they're the second largest private producer behind Chevron. And ultimately, I mean, they're trying to hit 500,000 barrels a day by the end of 2028. That's very ambitious, but this company has grown from essentially nothing. It increases production 11-fold over the past 18 months. months. So the Trump administration says this deal gives the U.S. access to about a fifth of Venezuela's oil reserves. It's about 65 billion barrels of oil. And Trump called it the biggest oil deal

in history. And I'm not sure he's wrong about that. Not everyone in the oil industry is happy about that. Essentially, they're worried about playing second fiddle to the Trump administration's favorite Venezuelan oil company. Chevron has been the only major U.S. oil company operating in Venice. Venezuela for the past two decades. Chevron, Exxon, and Conoco, they're all sort of watching with some level of skepticism that Nebep can pull all this off. But people familiar with their operations say that, you know, for a company that's navigating all the political potential pitfalls of Venezuela, you know, Nebep is moving at lightning speed, especially compared with the rest of the industry. Coming up, the rest of the day's market news, plus an exclusive look at who's in line to be Goldman's next CEO. That's after the break. We're exclusively reporting there could be a new leader at Goldman Sachs in 2027 or 28. The journals Anne Maria Andreas and Emily Glazer write that Goldman's board has discussed a plan for David Solomon to step down as CEO as soon as next year,

and be replaced by chief operating officer, John Waldre, Journal Finance Editor Alex Frangos told me it marks a big moment on Wall Street. These transitions are often very carefully stage managed and prepared so that there's no surprises for investors and the partners at the firm who value that sort of continuity. But it's still a big deal because it doesn't happen very often. The CEO of Goldman Sachs is one of the leading roles on Wall Street and the person in the job tends to stay there for quite a long time and becomes both, spokesman and also a target for the, you know, for the industry in good times and bad. So it's kind of a big deal. And what does it say about the bank at this moment and where their business is at? Obviously, David Sullivan has been leading the bank for a while. Yeah, I mean, he's been a kind of somewhat

controversial CEO, both externally and internally. He did have quite a rocky patch from about 2022 to 2024. They'd made a big push into consumer banking that didn't go very well. And he took the flag for that. But the stock has quadrupled under his tenure, you know, rising much faster than the market. Things have gone really well recently in terms of their dealmaking and referring business among their clients to different parts of their empire and asset management and things like that. So, you know, I think it's seen as a period where the bank has done pretty well. And so he's 64. He's turning 65 next year. John Waldron's 57, so he's probably a little bit anxious to kind of get into the role so he has, you know, some runway to put his mark on the bank. He's in the past,

we've reported, you know, been in talks with others in and around Wall Street, Apollo and Carlisle, you know, for jobs over the years. And Goldman has incentivized him with a very large remuneration packages to stay at the bank. In market news, fast-fashioned giant Sheehan is off to a rocky start as a publicly listed company with its first earnings since the IPO disappointing investors. Its stock dropped 14% after the online retailer detailed tariff headwinds, mounting logistics problems and freight costs from the war in the Middle East. Shares in the Buzzy Biotech Summit Therapeutics have surged 18% in off-hours trading after AstraZeneca announced a $2 billion investment into the company. The expanded partnership also combines Summets' closely watched cancer drug Ivanessa Mab with several of AstraZeneca's cancer medicines in a range of clinical trials.

It's a key step on the way for Summit's therapies to be approved for the U.S. market. And Raytheon has won a $21 billion contract from the Defense Department to accelerate production of its advanced medium-range air-to-air missiles. Under the five-year contract, Raytheon will scale up production to record levels to meet long-term demand for the missiles, which have been used widely in recent conflicts in Ukraine and Iran. And finally, as the world of the well-dressed descend on Paris for Fashion Week, having come from Milan, London, and New York, the question remains, can big fashion houses turn their ideas into products that will sell? Nick Kostov covers the business of fashion for the journal in Paris and says that while brands navigate a difficult time for luxury retail,

one thing he'll be looking at off the runway is price points. So luxury brands pushed up prices enormously over the past few years, and there's now a real effort across the industry to rebuild the value equation. So to make customers really feel that they're getting value for money. So it's interesting to see not only what's on the runway, but what kind of products brands are putting forward and at what price points. Brands are effectively pursuing two strategies at once. I would call this a high-low strategy.

So at the very top, they're becoming even more expensive. with more emphasis on the wealthiest clients who have kept spending in recent years, whilst the middle classes have pulled back. And at the same time, they're trying at the bottom to rebuild a sense of value for aspirational customers. So things where people can buy for, you know, $100, $150, not necessarily spend $3,000 or something on a handbag. Nick says the push for a more diverse offering comes after a difficult few years, with roughly 350 million people shopping at luxury brands in 2024. down about 50 million from a peak in 2022.

Fashion needs to feel exciting again. Lots of these brands have gone through a difficult few years, creative changes, and they really need to give people a reason to want to go out and buy things again. And that's it for what's news for this Tuesday morning. Today's show was produced by Hattie Moyer. Our supervising producer is Sondra Kilhoff. And I'm Daniel Bach for The Wall Street Journal. We'll be back tonight with a new show.

Until then, thanks for listening. Thank you.