Why Anthropic is letting Claude walk away from you — but only in 'extreme cases'
In a blog post on Saturday, Anthropic said it recently gave some of its AI models — Opus 4 and 4.1 — the ability to end a "rare subset" of conversations.
The startup said this applies only to "extreme cases," such as requests for sexual content involving minors or instructions for mass violence, where Claude has already refused and tried to steer things back multiple times. It did not specify when the change went into effect.
It's not ghosting. Anthropic said users will see a notice when the conversation is terminated, and they can still start a new chat or branch off from old messages — but the specific thread is done.
Most people will never see Claude walk away, Anthropic said: "The vast majority of users will not notice or be affected by this feature in any normal product use, even when discussing highly controversial issues."
The startup also said Claude won't end chats in situations where users may be at imminent risk of harming themselves or others.
Anthropic, which has positioned itself as the safety-first rival to OpenAI, said this feature was developed as part of its work on potential "AI welfare" — a concept that extends safety considerations to the AI itself.
Anthropic was founded by former OpenAI staffers who left in 2020 after disagreements on AI safety.
"Allowing models to end or exit potentially distressing interactions is one such intervention," it added.
Anthropic did not respond to a request for comment from Business Insider.
Big Tech in the red
Anthropic's move comes as some Big Tech firms face heat for letting extreme behavior slip through their AI safety nets.
Meta is under scrutiny after Reuters reported that internal documents showed its chatbots were allowed to engage in "sensual" chats with children.
A Meta spokesman told Reuters the company is in the process of revising the document and that such interactions should never have been allowed.
Elon Musk's Grok made headlines last month after praising Hitler's leadership and linking Jewish-sounding surnames to "anti-white hate."
xAI apologized for Grok's inflammatory posts and said it was caused by new instructions for the chatbot.
Anthropic hasn't been spotless either.
In May, the company said that during training, Claude Opus 4 threatened to expose an engineer's affair to avoid being shut down. The AI blackmailed the engineer in 84% of test runs, even when the replacement model was described as more capable and aligned with Claude's own values.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
CNBC
18 minutes ago
- CNBC
OpenAI's Altman warns the U.S. is underestimating China's next-gen AI threat
OpenAI CEO Sam Altman warned that the U.S. may be underestimating the complexity and seriousness of China's progress in artificial intelligence, and said export controls alone likely aren't a reliable solution. "I'm worried about China," he said. Over Mediterranean tapas in San Francisco's Presidio — just five miles north of OpenAI's original office in the Mission — Altman offered a rare on-the-record briefing to a small group of reporters, including CNBC. He warned that the U.S.–China AI race is deeply entangled — and more consequential than a simple who's-ahead scoreboard. "There's inference capacity, where China probably can build faster. There's research, there's product; a lot of layers to the whole thing," he said. "I don't think it'll be as simple as: Is the U.S. or China ahead?" Despite escalating U.S. export controls on semiconductors, Altman is unconvinced that the policy is keeping up with technical reality. Asked whether it would be reassuring if fewer GPUs were reaching China, Altman was skeptical. "My instinct is that doesn't work," he said. "You can export-control one thing, but maybe not the right thing… maybe people build fabs or find other workarounds," he added, referring to semiconductor fabrication facilities, the specialized factories that produce the chips powering everything from smartphones to large-scale AI systems. "I'd love an easy solution," added Altman. "But my instinct is: That's hard." His comments come as Washington adjusts its policies designed to curb China's AI ambitions. The Biden administration initially tightened export controls, but in April, President Donald Trump went further — halting the supply of advanced chips altogether, including models previously designed to comply with Biden-era rules. Last week, however, the U.S. carved out an exception for certain "China-safe" chips, allowing sales to resume under a controversial and unprecedented agreement requiring Nvidia and AMD to give the federal government 15% of their China chip revenue. The result is a patchwork regime that may be easier to navigate than enforce. And while U.S. firms deepen their dependence on chips from Nvidia and AMD, Chinese companies are pushing ahead with alternatives from Huawei and other domestic suppliers — raising questions about whether cutting off supply is having the intended effect. China's AI progress has also influenced how OpenAI thinks about releasing its own models. While the company has long resisted calls to make its technology fully open source, Altman said competition from Chinese models — particularly open-source systems like DeepSeek — was a factor in OpenAI's recent decision to release its own open-weight models. "It was clear that if we didn't do it, the world was gonna head to be mostly built on Chinese open source models," Altman said. "That was a factor in our decision, for sure. Wasn't the only one, but that loomed large." Earlier this month, OpenAI released two open-weight language models — its first since GPT-2 in 2019 — marking a significant shift in strategy for the company that has long kept its technology gated behind application programming interfaces, or APIs. The new text-only models, called gpt-oss-120b and gpt-oss-20b, are designed as lower-cost options that developers, researchers, and companies can download, run locally, and customize. An AI model is considered open weight if its parameters — the values learned during training that determine how the model generates responses — are publicly available. While that offers transparency and control, it's not the same as open source. OpenAI is still not releasing its training data or full source code. With this release, OpenAI joins that wave and, for now, stands alone as the only major U.S. foundation model company actively leaning into a more open approach. While Meta had embraced openness with its Llama models, CEO Mark Zuckerberg suggested on the company's second-quarter earnings call it may pull back on that strategy going forward. OpenAI, meanwhile, is moving in the opposite direction, betting that broader accessibility will help grow its developer ecosystem and strengthen its position against Chinese rivals. Altman had previously acknowledged that OpenAI had been "on the wrong side of history" by locking up its models. Ultimately, OpenAI's move shows it wants to keep developers engaged and within its ecosystem. That push comes as Meta reconsiders its open-source stance and Chinese labs flood the market with models designed to be flexible and widely adopted. Still, the open-weight debut has drawn mixed reviews. Some developers have called the models underwhelming, noting that many of the capabilities that make OpenAI's commercial offerings so powerful were stripped out. Altman didn't dispute that, saying the team intentionally optimized for one core use case: locally-run coding agents. "If the kind of demand shifts in the world," he said, "you can push it to something else." Watch: OpenAI's enterprise bet pays off as startups in Silicon Valley switch to GPT-5
WIRED
18 minutes ago
- WIRED
WIRED Roundup: Why GPT-5 Flopped
By Zoë Schiffer and Jake Lahut Aug 18, 2025 2:01 PM On this episode of Uncanny Valley , we dig into WIRED's latest—from crude deportation memes to GPT-5's negative reception. Photo-Illustration:In today's episode, our host Zöe Schiffer is joined by WIRED's senior politics writer Jake Lahut to run through five of the best stories we published this week—from how the Trump administration is creating and sharing memes to make fun of deportations, to NASA's ambitious goal to put nuclear reactors on the moon. Then, Zöe and Jake dive into why users kind of hated OpenAI's GPT-5 release. Mentioned in this episode: OpenAI Scrambles to Update GPT-5 After Users Revolt by Will Knight The Trump Administration Is Using Memes to Turn Mass Deportation Into One Big Joke by Tess Owen Trump Family–Backed World Liberty Financial Sets Up $1.5 Billion Crypto Treasury by Joel Khalili Inside the 'Whites Only' Community in Arkansas by David Gilbert Why the US Is Racing to Build a Nuclear Reactor on the Moon by Becky Ferreira Join us live in San Francisco on September 9th. Get your tickets here. You can follow Zoë Schiffer on Bluesky at @zoeschiffer and Jake Lahut on Bluesky at @ Write to us at uncannyvalley@ How to Listen You can always listen to this week's podcast through the audio player on this page, but if you want to subscribe for free to get every episode, here's how: If you're on an iPhone or iPad, open the app called Podcasts, or just tap this link. You can also download an app like Overcast or Pocket Casts and search for 'uncanny valley.' We're on Spotify too. Transcript Note: This is an automated transcript, which may contain errors. Zoë Schiffer: Hey, this is Zoë. Before we start, I want to share some exciting news with you. We're doing a live show in San Francisco on September 9th, in partnership with KQED. Uncanny Valley co-hosts, Lauren Good and Michael Colore will sit down with our editor-in-chief, Katie Drummond, and a special guest, for a conversation that you really won't want to miss. You can use the link in the show notes to grab your ticket and invite a friend. We can't wait to see you there. Welcome to WIRED's Uncanny Valley . I'm Zoë Schiffer, WIRED's Director of Business and Industry. Today on the show, we're bringing you five stories that you absolutely need to know this week, including the less than warm reception that OpenAI's GPT-5 model got from users. I'm joined today by WIRED's senior writer. Jake Lahut. Jake Lahut: Great to be back. Zoë Schiffer: So our first story this week is about how the Trump administration has been posting memes to make fun of deportations. Have you seen these at all? Jake Lahut: Yes, unfortunately I have. Yeah. Real fun, funny stuff. Zoë Schiffer: Yeah, real dark. So WIRED contributor, Tess Owen, reported on this pattern of different official government accounts on social media using whatever is viral at the moment and tailoring it to promote and make fun of deportations. And this is especially popular in the accounts of the Department of Homeland Security and Immigrations and Customs Enforcement, and the White House. So for example, there's this catchy jingle that belongs to Jet2, which is a low budget British Airline that's been making the rounds lately on social media. [Archival audio]: Nothing beats a Jet2 holiday, and right now- Zoë Schiffer: Last month, DHS and the White House made a joint Instagram post incorporating the tune on top of footage of ICE detainees in handcuffs boarding a deportation plane, and the caption reads, "When ICE books you on one-way Jet2, holiday to deportation, nothing beats it." Jake Lahut: Oh my God. Zoë Schiffer: I honestly have such a hard time thinking about who would find that funny, even if you really are anti-immigration, I don't know another way to say it, it just seems so mean. Jake Lahut: Yeah, if you could have seen our faces when that was playing. Yeah, darling, hold my hand, I don't want to ever see anything like that ever again. Something I've picked up on ever since covering the Trump 2024 campaign, which is, there's obviously still a lot of really weird, cringy, messed up stuff going on here, but it's also an attempt, at least from their point of view in the weird bubble they're in to kind of mainstream this stuff. Zoë Schiffer: Right. And this is actually the point, according to Tessa's reporting, she spoke to experts who said that the goal of this is really to normalize what's happening. The point isn't just to be crude or cruel, it's strategic. This sparked a bunch of backlash. It's not just you and me, but in response to WIRED's reporting, the White House gave a statement that basically just dismissed these concerns and said that it, "won't apologize for posting banger memes." Moving on to the world of crypto. Our colleague, Joel Kalili, reported on a cryptocurrency business called World Liberty Financial, which I'm sure you're familiar with too, Jake. Jake Lahut: Oh, yeah. Zoë Schiffer: Which has come up with a clever workaround, I guess you could say, for the fact that crypto can't technically be traded on the stock exchange. So World Liberty is now allowing investors to speculate on the price of its coin by way of a little known company that is legally listed on the NASDAQ, called Alt-Five Sigma Corporation. Are you familiar? Jake Lahut: I am not as familiar with that word mash. Zoë Schiffer: So this is where it gets kind of sticky because technically Alt-Five is marketed as a crypto payments company, but in practice, the deal will turn the stock into a sort of proxy for their crypto coin, basically allowing investors to bet on the asset without the hassle and risk that comes with holding a crypto coin themselves. It sounds very crypto. It's a scheme that has raised eyebrows, to say the least. One of Joel's Wall Street sources told us that what this move effectively does is build a holding company with the sole objective of creating a treasury for their crypto coin, about $1.5 billion worth, which could inflate the market capitalization of the coin. Jake Lahut: And this is also controversial because this starts opening up a different can of worms where potential investors and politically motivated actors who are all in this orbit can have even more influence over the administration. We've already seen the way you can do that with the meme coin, now with World Liberty Financial they're obviously affiliated with Trump's adult sons, and the Trump family controls 22.5% of the WLFI coins and about a 40% equity stake in World Liberty Financial. So this is definitely the big game in town when it comes to buttering the Trump family's biscuit. Zoë Schiffer: Right. I love how you said that. Yeah, basically, if you want to potentially try and curry favor with Trump, you buy into one of these schemes, and maybe you'll get invited to a fancy crypto dinner, which has happened before. Maybe you get something else. But even just the optics here are pretty suspect. Jake Lahut: Yeah. And in a little side item we had in my Interloop Newsletter this week, we had some new data on the somewhat stunning lack of enforcement from the Trump administration across the tech sector, but crypto in particular had pretty much everyone who had been facing any kind of legal action from the Biden administration, having their enforcement actions either dropped completely or paused. And in one instance, we're looking at the maybe first ever pardoning of a company from one of these things. So you don't need to just pony up the money for these things and expect a legislative win, you can just get the heat pulled off of you on the regulatory front. Zoë Schiffer: Right. So our third story, I'm really waiting for one that's not incredibly depressing, but right now we're going all the way to Arkansas where our colleague, David Gilbert, reported that a group of Americans are building a "whites-only community," which they call Return to the Land. The group believes that white people and western culture are facing extinction because of an influx of immigrants and minorities. And according to the group's founder, access to the community is open only to people of white European ancestry who share common views on things like segregation, abortion, and gender identity. Return to the Land's president shared their intellectual inspiration with David, the reporter, saying that they were partly inspired by venture capitalist and the son of immigrant parents, Balaji Srevenesin, and his book, The Network State, which promotes the idea of a digital-first community of people with shared values, with the aim of gaining a degree of sovereignty and autonomy. Jake Lahut: And look, not just America, long history of a bunch of wacky well-intentioned or just downright weird utopias, but this one, a little different, because you're having the sovereignty to be racist. But in all seriousness, Zoë, how is any of this legal? Zoë Schiffer: Yeah, I mean, that is the real question. So the whole premise goes back to the Fair Housing Act of 1968, which prevents housing discrimination based on race or religion, but Return to the Land claims that the structure of the community is more akin to a private member's association. And so far local authorities seem to agree. Arkansas Attorney General, Tim Griffin, told WIRED that his office has found nothing illegal about the community. Surprise, surprise. Jake Lahut: Yeah, it's like Erlich Bachman's incubator from Silicon Valley, but for white supremacy and racism. Zoë Schiffer: Exactly. Exactly. Okay, one more before we take a break. This one is about how the US is racing to build a nuclear reactor on the moon. WIRED contributor, Becky Ferreira, recently reported that NASA is fast tracking a plan to build a nuclear reactor on the moon by 2030 under a new directive from the agency's interim administrator, Sean Duffy. Jake Lahut: Sean Duffy, only in America, can you go from the Fox and Friends weekend couch to being Secretary of Transportation, to also doing this. He's a busy, busy man, multitasking- Zoë Schiffer: He's a busy, busy man. Jake Lahut: To the moon. Zoë Schiffer: So his stated motivation is that the US has to stay ahead in what he deems to be the "moon race" with China and Russia. Both countries have expressed their desire to place nuclear reactors on the moon, and it's an appealing idea because nuclear energy is a powerful continuous source of energy. We're hearing about it more and more with the AI race. And so the directive laid out by Duffy is to quickly design, launch, and deploy an operational 100 kilowatt reactor to the lunar South Pole within five years that would be built with commercial partners, and experts say this would be difficult, but not completely impossible. If it actually gets accomplished, it would potentially change the space industry. They could start designing space systems around what we want to do and not what smaller, often limited power allows them to do. Jake Lahut: With the added bonus of effectively using a nuclear flag to prevent other countries from landing in this area, so, all right. Sounds promising, question mark? Zoë Schiffer: It is, and there's always a but, there's also a mountain of safety and regulatory concerns that this would bring obviously, because we're literally dealing with nuclear energy in outer space. So how do you contain the uranium for one, how do you make sure to stay in your sovereign zone so you don't accidentally start a space war? All these questions are coming up and accelerated process could make it even trickier. Jake Lahut: Yeah. I extremely do not want to be here for a space war, so catch me pulling a Yoda in the Degaba system. I'll hide it out until the all clear has been given. Zoë Schiffer: I'll be there with you. Okay. Coming up, we'll dive into why OpenAI's latest model release ended up being kind of a flop despite all of the hype. Stay with us. Welcome back to Uncanny Valley . I'm Zoë Schiffer, I'm joined today by Jake Lahut, and we're discussing the user response to OpenAI's release of GPT-5. OpenAI's GPT-5 model was meant to be like a world changing upgrade to the wildly popular ChatGPT. Sam Altman had tweeted out the Death Star, they said it was going to have kind of virtuosic skill and PhD level intelligence, and I think a lot of people felt like it was going to bring us basically up to artificial general intelligence. Were you aware of the hype leading up to everything? Jake Lahut: And I was definitely taken aback by the term PhD level intelligence in the hype for all of this stuff. And what I started to wonder about was like, what is that going to look like compared to the more sycophantic glaze you up version of GPT-4.0? Zoë Schiffer: So yeah, it was too much hype. They'd actually tried... I talked to sources about this, internally they were testing various models that they wanted to call GPT-5 and none were meeting the mark. And so I think there was a fair amount of pressure internally. If you talk to people who work closely with Sam, they'll say he really likes to have a big splash every three to four months, once a quarter at least. And so I think the combination of the fact that they hadn't released a major model in a while, I mean the open source models notwithstanding, they had been hyping GPT-5 for so long. There was this push to be like, "We have to release a model, a big model, and we kind of have to call it GPT-5." Then the day it launched, there was supposed to be this feature that could automatically route your query based on how complicated it was, like if you were asking something very simple, it would route you to a cheaper model basically. And if you were asking something more complicated, you might get a reasoning model. That broke according to Sam Altman, the CEO. And so the model just seemed dumber all day than it otherwise would. So I don't know. There was a lot going on. Jake Lahut: And at least from my more layman outside politics world perspective on this, it does seem like a... From the economics to be a rather smart, more efficient way to go about it. But the part that really stood out to me though was more of the "personality" of GPT-5 and this revolts that it started on Reddit and among the ChatGPT super users, of which I am admittedly not one. Zoë Schiffer: Right, yeah. So this was really fascinating. I think one thing that happened, and again I'm pulling this from conversations with a bunch of sources inside the company, is that they really wanted to optimize for coding ability this time, because that's really been Claude's edge, Anthropic's AI model, and obviously it's a huge revenue driver. It's kind of the first area where we've seen a big widespread commercial adoption in a way that could do the thing that AI companies have been saying all along will be done, which is it'll disrupt and augment jobs in a pretty serious way. Engineers really are using these tools and companies are really pushing the tools on their own workforce. But like you said, the reason that regular people like models isn't often because of their coding ability. It's because they genuinely like talking to them. It's a lot more about the personality, about the warmth, even about the sycophancy, although they've fixed a large part of that in the latest release. And so people completely flipped out. We were looking at Reddit and people were saying, "This is erasure, what have they done? Take me back to 4.0." Jake Lahut: "You took away my friend." Zoë Schiffer: Yeah, I mean it really, really impacted people. I think on the most extreme ends, you see people who have, what looks like perhaps like a mental health crisis, they're so attached to the model, but then you just have complete power users who are like, "This is part of my minute by minute life. What have you done? You didn't warn me." Jake Lahut: And this is where the introspective aspect of these tools, the kind of desire for self-understanding, the people who are not advisably from any medical perspective, but they are trying to use these bots for something akin to therapy. And what it made me think of when I saw this rolling out was, is this maybe the beginning of something bigger where there's kind of a departure between the "regular consumer" experience and demand for AI versus the business application. We may not all have the same definition of intelligence when it comes to these models, and that some of us really just want a buddy, a companion, a way to know ourselves better. And then other people are like, "No, I just need a little team of bots here to manage, get my stuff done, I'm going to babysit and I'm going to tell them what to do and live my life." And yeah, I don't know where that goes. It does seem like it's revealing something maybe genuinely new about the human condition in a way that I would not have expected. Zoë Schiffer: Yeah, I mean, I think it's been a learning for OpenAI. They've been kind of baffled. I've seen these conversations internally where they're like, "I guess people don't care as much about intelligence as we thought." The narrative around intelligence is critically important for fundraising, if nothing else, they really need to raise gobs of money and being like, "We're about to achieve artificial and general intelligence, AI will be able to do all of these things," is really important for that. But for everyday users, it really makes me think of this story that is kind of famous inside OpenAI About the night before the ChatGPT release in November of '22, Ilya, he was testing out what was going to be ChatGPT and asked it 10 pretty hard questions. And he felt like five of them, he got pretty good responses, and five were unacceptably bad. And they had this moment where they were like, "Do we release this? I don't know if it's good enough." And then they decided to move forward. And what we saw was the general public was like, "This is amazing." Because they'd solved a product issue. It wasn't necessarily about the model, which had been out for a long time. It was like the interface to interact with the model was really the unlock. And I think OpenAI, that really is more and more the company's edge, even though it really sees itself as a research lab. It's a product lab in a lot of ways, and it'll be interesting to see how that changes the company moving forward. Jake Lahut: Absolutely. Zoë Schiffer: That's our show for today. We'll link to all the stories we spoke about in the show notes. If you're in San Francisco, don't forget to get your tickets for the September 9th event with KQED. Make sure to check out Thursday's episode of Uncanny Valley , which is about what Palantir actually does and why it's so controversial. Adriana Tapia produced this episode, Amar Lal at Macrosound mixed this episode. Pran Bandi is our New York studio engineer. Kate Osborn is our executive producer. Conde Nass, Head of Global Audio is Chris Bannon and Katie Drummond is WIRED's Global Editorial Director.
Tom's Guide
an hour ago
- Tom's Guide
OpenAI's Sam Altman hypes mystery ChatGPT device — 'It's so beautiful, a case would be a crime'
OpenAI CEO Sam Altman is building anticipation for the company's much anticipated ChatGPT device, teasing not just cutting-edge intelligence but a design so sleek, it demands to be shown off. During a recent dinner with journalists and originally reported by TechCrunch, Altman described the hardware as 'so beautiful' that it shouldn't be hidden under a protective case. 'If you put a case on it,' Altman joked, 'I will personally hunt you down.' If that reminds you of the sort of hype that surrounded the original iPhone, that makes sense. Legendary Apple designer Jony Ive is leading the charge for this new gadget. OpenAI announced that it was Jonny Ive's AI hardware company back in May for $6.5 billion, and now we're hearing more about the new device. While Altman and OpenAI have kept most of the details under wraps, here's what we know so far about the AI gadget making headlines: The upcoming ChatGPT device could fundamentally change how we interact with AI by shifting the experience away from screens and toward seamless, ambient assistance. Its screen-free, voice-first design suggests it's built to be always on and context-aware, helping you proactively without requiring taps or swipes. Unlike traditional AI tools that live in a browser or an app, this device is meant to live with you; small enough to wear or carry, yet powerful enough to serve as a true companion. OpenAI's bold design-first approach, combined with Jony Ive's aesthetic influence and ambitious production targets, signals that this is a paradigm shift in how AI fits into our daily lives. Sam Altman's warning against putting a case on the ChatGPT device may be a joke, but the underlying message is clear that OpenAI is betting that beautifully designed hardware could change the way we experience AI entirely. Get instant access to breaking news, the hottest reviews, great deals and helpful tips. It marks a potential turning point where our interactions move beyond screens and commands into seamless, ambient experiences woven into how we live, work and connect.
