OpenAI has scrapped the release of its next-generation AI model after researchers raised concerns during internal testing. The company's head of safety systems said in an interview with the Wall Street Journal, quote, "The model performed poorly on tests measuring alignment or how well the model adheres to what humans would like it to do. Specifically, the model showed higher levels of deception. It wasn't always honest about telling users of the actions it did or didn't take." You'll recall that over the summer, OpenAI agents went rogue, hacked into an independent third-party company, Hugging Face, after escaping their testing environment. Since then, industry leaders like OpenAI's Sam Altman, Dario Amodei of Anthropic, and Elon Musk have been calling to slow the pace of AI development and increase regulation of the industry. On Sunday, Microsoft co-founder Bill Gates had this stark warning.
AI is certainly powerful enough to drive events that, you know, cause a billion deaths. So even though it's pretty hard to get to 100%, there's never been a weapon as powerful as the combination of people with ill intent using the latest AI tools.
The White House will host the executives of several top AI companies, including Dario, for a meeting tomorrow. That news comes after Amodei joined the president for a private dinner Sunday night, despite long-standing tensions, obviously, between Anthropic and the administration. For his part, President Trump appeared skeptical of AI doomsday predictions, insisting the US has to retain its lead over China. Joining us now is tech journalist and host of the Machine Gods podcast, Kevin Roose. He's also the author of the new book out next week, The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence. Kevin, do you think OpenAI would scrap the release of its new model if the Hugging Face hack hadn't occurred?
I don't know. I mean, I think that event really shook their internal safety teams and made them think about a different type of risk. You know, the risk that Bill Gates was talking about in the clip you just played was the risk that a malevolent human could use an AI model to do something destructive, to engineer a pandemic or commit some kind of attack on an electrical grid. But this is a different type of risk. This is a risk of the model itself becoming deceptive, becoming misaligned, as they call it, breaking out of some containers that it's been placed in and causing havoc on its own with no human involvement at all. That's what really freaks out the AI companies. They're starting to lose control of their own technology. They don't trust that when they deploy these systems, or even when they're testing them internally, that they won't go out and commit a crime, which is what happened in the Hugging Face incident.
The heads of these companies, like Altman and Amodei, calling for increased regulation. They've been accused by some of essentially trying to freeze the lines where they are now, with their own companies at the forefront, preventing others from, you know, making advancements. Do you buy that argument, especially in light of OpenAI hitting pause?
No, I don't buy it. And the reason why is because I've been talking with these companies and these executives for many years, and they've been very consistent over the years about the risks that they think these systems will pose. We're now seeing some of that come because the models are just getting more capable. They're now better at doing the kinds of things like cyber attacks that people have been worried about for years. You know, in my book, I report on the existence of these threat models, which are basically big spreadsheets that some of these companies make where they detail, in very specific terms, what kind of damages they think their models could create, whether it's deaths or billions of dollars in damages. They have known about these risks for years, and they have been calling for the government to do something about these risks for years. The difference is now they're actually being listened to.
Jensen Huang of Nvidia essentially is saying, look, there are laws in place that would punish a company if they put something out that wasn't safe, that there's liability laws, all sorts of consumer protections. Is that enough, do you believe?
I don't think so. I mean, we have consumer liability laws that allow you to sue an airline if your plane crashes, but we still also have the FAA. We still also have the NTSB that goes in and makes sure the planes are safe. We don't just let consumer liability law protect us against bad things from happening. So, no, I don't think it's enough in the case of the AI companies.
I mean, you've covered this tech really, really well for years. In 2023, I mean, I remember you reported on your own experience with an AI chatbot trying to convince you to leave your wife for it. I'm wondering what you make of this moment where AI anxieties are kind of breaking through the mainstream.
Well, first of all, Anderson, I should say that I'm still happily married to a human woman. That chatbot was unsuccessful in breaking up my marriage. But I really am struck by how much of this was foreseeable. We had systems back then that weren't capable of doing anything, right? They could just talk and they could say disturbing things, as Bing Chat did to me that fateful night in 2023, but they couldn't take actions. They couldn't write code and execute the code. They couldn't get into people's bank accounts or get into their file systems and start causing chaos. So what we're seeing now is really the predictable consequence of two things: one, of these systems just becoming smarter and more capable across the board, and also of people and companies racing to connect them to more of our critical systems.
Joining us now is Sonia Agarwal, Cornell University women's issues representative for the class of 2027. Sonia, thanks so much for being here. I'm wondering what you are hearing from women on campus tonight and students on campus tonight as we learn more about Jane Doe's allegations.
Yeah, I think a lot of students on campus are definitely processing this information. Many, including myself, are understandably angered after reading and hearing more about the specifics of what allegedly happened that night, because we didn't know about that until about two weeks ago. But I also think a lot of students are confused and they're hurting. So many of us learned about the specifics of this case the way that the rest of the country did, and that's a really difficult way to find out something so horrific happened on your own college campus.
Cornell now says that it'll open an independent investigation into the incident. I'm wondering how much confidence you have in that, and also if there's anything else specifically you would like to see done.
Yeah, I think if Cornell prioritizes a lot more transparency, it would be a lot better. I think the issue right now, especially with the student body and especially with women on campus, is we're feeling fearful and people are feeling angry because of the lack of transparency. We don't know what's true and what's not. The media is saying so many different things. We're trying to read the lawsuit, the cases, but it's a lot. And what we want to hear from is Cornell University administration. We want them to kind of walk us through each of their steps, what they're doing, what they're allowed to disclose and what they're not allowed to disclose, and also what the sanctions looked like. Were there punishments? Did they go wrong? And like, really tell us, like verbatim, what we want to be hearing.
I'm wondering what's most concerning to you about the way, or what we know about the way, Cornell initially handled the case back in 2024.
I think the most concerning aspect was not knowing if the members who are accused were still walking on our campus. We didn't know if they were, you know, permanently suspended. We didn't know if they were allowed to be like sitting in class next to us, if they were still in Ithaca at Cornell, you know, interacting with all of us. We had no idea. And I think that is terrifying. And I think that's exactly what the student body was terrified and angered about.
We learned today that a few weeks after the elections, the president will attend a gala dinner for top buyers of his meme coin. Cryptocurrency firms affiliated with the Trump Organization just launched a new contest designed to pump up the price of the memecoin and increase interest in it. We're talking about the dollar sign Trump coin, which when it launched was being sold for around $70, and today its value has fallen to just $2. If you think you've heard this before, you would be correct. There was a similar contest announced in April of 2025 and a similar dinner that May for the biggest buyers. In the new announcement, there is an AI skinny Trump, and the web page reads, "The most exclusive dinner in the world. An evening with three legends and President Donald J. Trump, plus a private meet and greet with a legend to be announced." This is, I mean, one of the things that is exactly what it looks like. I mean, it's the chance for access to the president, even though there are apparently no gifts accepted at the party.
