AI is the hottest thing in medical care right now — but many of us feel trepidation about it. Just one illustrative public survey sample: An October 2025 KFF poll found just 8 percent of Americans reported feeling a “great deal” of trust in AI managing their appointments or analyzing their health records, and only 32 percent said they would trust an online health tool that uses AI to access their medical records to provide personalized health information.

But many clinicians and healthcare administrators see AI as a powerful new tool that offers myriad opportunities to streamline and improve treatment. A 2026 survey found that more than 80 percent of US doctors use AI professionally — doubling the share from 2023. Physicians are excited by AI’s potential to keep more accurate notes of interactions with patients, to act as a second pair of eyes for human doctors, and to monitor people at risk of deteriorating and ending up in a dangerous situation.

The disconnect between what people and their providers want from AI could create more distrust, at a time when faith in the healthcare system and the medical profession have slid. Patients today want to feel empowered and in control. How can that be possible when these seemingly godlike machines are becoming more and more entrenched in our hospitals and doctors offices?

The answer comes in four words: “human in the loop.” It’s the principle upon which the ethical integration of AI depends and it could help to bridge the gap between lay people and the professionals on AI in medicine. In surveys, people are much more comfortable with the idea of their doctor using AI as an assistant than with AI acting on its own. And most clinicians want to use AI in that way, as a second opinion or passive monitor, not as a replacement for their judgment. There are real fears among the healthcare workforce about that possibility: A group of NYC nurses who were recently laid off claim it’s because their labor was going to be replaced by AI. “Human in the loop” appears to be a point of agreement between doctors and patients at this pivotal moment.

“Doctors…and nurses and staff always have been interested in primarily making the best decision for the people under their care — and these tools can help with that,” Alison Callahan, a research scientist at Stanford University who works on AI programs used in the university’s health system, told me. “The interest in making sure those tools are accurate is high.”

But what does “human in the loop” really mean in practice? How can you know when and how your doctor is using AI? And what is the best way to talk to your provider about the sudden influx of artificial intelligence in healthcare before a robot starts taking appointment notes or analyzing your MRI? I called some leading experts to find out. 

How AI is currently being used in medicine

Patients and providers alike are incorporating AI into healthcare. Individuals are using commercial AI chatbots to ask about their symptoms or the health metrics tracked by their Apple Watch, while large academic medical centers are developing sophisticated programs and protocols to try to improve medical care at the population level.

It starts with ChatGPT, Claude, etc. — the large language models that are available to the public. People are increasingly turning to them to try to understand what’s going on with their own bodies. Individual physicians are also consulting with large language models to answer questions or get up-to-date on the latest research as they figure out how to best care for their patients. 

Then there are ways in which hospitals and doctors offices are adopting AI at the institutional level. Many facilities are using AI as a way to take, collate, and summarize notes on a patient; in theory, it’s a more organized way to keep track of the informal interactions and observations that doctors have when checking on their own patients. Hospitals are also using AI to handle some administrative tasks, like scheduling follow-up appointments; some health systems have even started to use AI to help patients get ready for appointments — to send reminders about colonoscopy prep, for example.

And finally, you have maybe the most ambitious use of AI by health systems right now: as a diagnostic and risk prediction tool. In these cases, AI might offer a second opinion when, for example, a doctor is triaging a patient in the emergency room. It might help the ER staff figure out how to prioritize patients. Or these programs could monitor people either during a hospital stay or out in the real world (by drawing data from the person’s wearable) and make predictions about who may be at higher risk of complications and require further care. AI could recommend that somebody would benefit from seeing certain specialists or receiving a specific medicine or lab test, and generally offer proactive advice about the patient’s medical care.

But at this point, AI adoption is still “highly localized,” said Jennifer Goldsack, CEO of the Digital Medicine Society, a nonprofit that works with healthcare providers, drug makers, and government agencies on how to incorporate new tech (including AI) into clinical care. It depends on the individual doctor or health system. A lot of them are setting up their own programs and their own protocols for how to use these tools.

That is a big reason why it is so important for patients to be proactive about understanding how AI is being used for their health care. You can’t make assumptions; the only way you’re going to know for sure is to ask.

The questions you should ask your doctor about AI

By and large, experts say, patients should feel confident: Doctors and nurses want to keep a human in the loop, even as they integrate AI into their workflows.

“It will be a doctor who is going to be reading that summary or a nurse who is going to be reading that summary and then taking an action to order a lab or put a recommendation in for a follow-up appointment,” Callahan said. “There is high interest in making sure that that is the right decision for that person. That hasn’t changed.” 

Still, many patients say they’d be more comfortable with AI use if their doctor fully explained it in advance. And health systems may have their own priorities that push their facilities toward more rapid AI adoption and delegating more tasks to these AI tools, as seen in the recent NYC nurse layoffs.

So if you want to be informed on exactly where this technology is present and have the ability to consent to its use, you have every right to ask your doctor, experts say. 

“AI is new, but the trust that serves as the foundation of the physician-patient relationship is not,” Timothy Keyes, a machine learning scientist at Stanford Health Care, told me over email. “To that end, I think that conversations about medical AI use should be open, honest, and transparent — just like any other conversations about shared decision-making in the clinical environment should be.”

For some things, your doctor should be asking you proactively if you consent to AI use — note-taking, for example. At my most recent primary care appointment, my doctor asked me if it’d be okay for him to use AI to take and summarize notes from our conversation; Goldstack told me she’d experienced the same at recent physician visits. (This is probably the most common AI use that you will encounter, and Keyes said it’s worth considering giving your consent: “There is growing evidence that they reduce physician burnout and save them at least a bit of time each day writing notes.”)

There are also a number of direct questions that you can ask:

• Will AI be used in my care and how?
• How is my data being protected?
• Can I opt out of any AI services that I do not feel comfortable with? (Keyes noted that patients should be allowed to opt out of any care, AI-related or not; if opting out is not an option, ask how a human provider will be involved.)
• How is the health system or clinic making sure that any AI system they use is working as intended?

And the transparency goes both ways. If you’re asking a question because you consulted ChatGPT before your appointment, tell your doctor. If you’ve talked with a chatbot because of mental health struggles, tell your doctor. And at the same time, feel free to ask your physician how you yourself could actually use AI in a responsible and productive way to improve your health.

“This opens up the opportunity for both the physician and the patient to be humans-in-the-loop,” Keyes said, “in different parts of the loop, with different perspectives, using an AI system to better understand the bigger picture.”

In a way, the novelty of AI and its rapid adoption is an opportunity for all of us to be nosier and more inquisitive patients. What all of these questions really come down to, Callahan said, is how your doctor is making decisions about your health care. That is relevant to all of us, no matter how AI is involved or even if there is no AI being used at all. 

Callahan said she always has a list of questions for her doctor when they recommend a course of treatment: “What are the factors in my health that are informing this recommendation that you have? Would you be making this recommendation for other patients who are similar to me? What can you tell me about the outcomes that I might expect to experience if I say yes to this?”

“I actually think if they can point to the part of your health that is connected to the decision, whether or not an AI tool helped to make that connection is secondary to their ability to communicate effectively to me about it, and help me to feel engaged in making a decision about my own care,” she said.

AI is changing medicine quickly, for both patients and their doctors. The best way to stay ahead is to talk about it.

It’s the act that’s defined a generation, an experience so common it’s become memeified: canceling plans. Bailing is a litmus test of sorts; who and what events you back out of says a lot about you as a person. Are you the type who commits and only bails when absolutely necessary? Or are you a frequent flake looking to change their ways? (No judgment here: The occasional canceling of plans is acceptable, and you shouldn’t feel guilty about it.)

Call it a byproduct of overscheduling, a facet of introversion, or a condition of mild social anxiety, but bailing is a popular tool in our social arsenal. Regularly deployed mere hours ahead of a previously scheduled engagement we’ve come to dread, canceling was once considered a get-out-of-obligation-free card. In recent years, though, general consensus around bowing out has seemed to shift, perhaps spurred by the growing attention paid to the dangers of social isolation or simply because being consistently bailed on feels like garbage. All of which is to say, canceling requires much more tact than simply rattling off a text 30 minutes before you’re supposed to meet up with a friend.

There are, of course, valid reasons to skip a movie night (you’re sick, for instance) and very few excuses good enough to justify not showing up to a wedding you RSVP’d to. But rather than giving in to the impulse to flake when the urge hits, it’s important to make a more considered choice based not on how much you think you’ll enjoy a social event but on the consequences of skipping. 

Consider this your canceling checklist, a bailing thought experiment, a toolkit for bowing out. Taken together, the following five questions will help inform your decision on whether to skip a social engagement.

Who are you bailing on?

Before flaking, take your audience into consideration — namely, whether this is a person you’d be comfortable inconveniencing. Because we’ve already invested significant time and energy into close relationships with family and friends, we’re more likely to want to preserve those relationships by upholding plans. This past history can influence how comfortable we are with bailing.

“How important is this person?” Julian Givi, an associate professor of marketing at West Virginia University, tells Vox. “Do I value this person’s friendship, or are they someone that I’m okay with canceling on? Have they canceled on me?”

Close friends are complicated. In a study, William Chopik, an associate professor in the Department of Psychological and Brain Sciences at Washington University in St. Louis, found that participants reported being most upset when a best friend canceled on them, compared with when a good friend or acquaintance canceled. Ironically, though, that closeness gives us latitude. People also tend to be more forgiving when close friends cancel on them, Chopik says. By virtue of knowing someone so intimately, we understand the constraints on their time and can intuit they’d only cancel for a good reason. It’s also safe to assume we’ll likely see them again soon, too.

We may feel less obligated to those we don’t know well. The participants in Chopik’s study said they’d only be “a little” upset if an acquaintance canceled plans. So, if this is someone you don’t know very well, and don’t necessarily plan to, you might feel more comfortable ditching them.

What’s the occasion? 

A monthly game night and a funeral don’t carry the same significance. You can probably guess which one is acceptable to skip. The formality of the event will give you a signal on whether it’s alright to bail. “People are more likely to cancel informal things,” Chopik says.

It’s fine to bail on a coffee hang, but think twice before ditching a wedding, funeral, retirement party, or other event of importance where someone might’ve paid money for you to be there, even if you don’t necessarily deem it a monumental occasion. “If it’s super important to this person, I’ll go, I’ll move mountains, I’ll traverse seas, I’ll travel as much as possible or inconvenience myself a lot,” Chopik says. “But then if it’s more informal or you don’t know them very well and the burden to get there is high, those things are the things that’ll lead you to likely cancel.”

While it may seem acceptable to ghost on a group occasion — who would notice you’re missing anyway? — your absence might alter the vibe. Take a bachelor or bachelorette party: If five of seven invitees back out, the celebration has shifted from a group affair into an intimate gathering. “Those last two left are in such a weird situation where it’s like, now I don’t want to go as much because it’s a smaller bachelor party,” Givi says. Your bailing signals to the stragglers that the event isn’t fun enough for everyone to attend.

How annoying is it for you to attend?

How often do we schedule a hangout only to remember later how much of a burden it’ll be to get there? The costs — financial, logistical, and energetic — all factor heavily into which plans we see through. A long and expensive commute for a short, casual hang when you’re already exhausted may not seem worth it. 

The problem is, we’re bad at predicting how much fun we’ll have. We tend to overestimate the costs and underestimate how enjoyable the event might be, which leads us to cancel plans. In reality, our worst fears almost never materialize. “If you’re really anxious about it, it’s probably not going to be as bad as you expect. And that’s a pretty consistent finding,” Ella Moeck, a lecturer at Adelaide University in South Australia, tells Vox.

Everyone’s bar for what constitutes “too far” or “too much effort” varies, though. Your canceling over a 15-minute drive across town runs the risk of pissing off a friend who doesn’t consider that a hassle at all. 

Concerning money, when people have already paid up front, they’re more likely to stick with plans. In that case, canceling would create more problems instead of quelling them. “That’s why anytime I’m planning anything, especially if it involves money, I make people Venmo me weeks in advance,” Givi says, for, say, a show or concert. 

What’s your excuse?

In his study on canceling plans, Chopik found the most reasonable excuses for canceling were being sick, dealing with a family emergency or childcare issue, or having a work obligation. The worst justifications were mentioning something better came up, not being interested in the activity, and straight-up lying. 

Surprisingly, participants weren’t sympathetic when someone canceled because of financial constraints. Chopik suspects the framing matters more than anything else: “I bet if you communicated that ‘I would’ve loved to go to this concert and spend time with you…but I just really can’t squeeze this right now,’ that’s a way more understanding excuse than being like, ‘That’s too expensive. I’m not going,’” he says.

The excuses themselves carry weight, but what they signal is also crucial. By telling a friend you’re bailing on them because you’d rather hang out with someone else, you’re communicating that they might not matter to you. But if you provide clarity about why you can’t make it and try to reschedule, that’s a sign you value the relationship. “If it’s just like, ‘I’m tired,’ come on, you committed already,” Givi says.

What are the consequences?

Canceling plans comes with potential fallout. Bowing out of a recurring group event isn’t likely to ruffle many feathers, but sending a “Sorry, can’t make it!” text minutes before a dinner reservation with out-of-town family members might not fly. Think about the position you’d leave the other party in, Givi says. “You could imagine it’s probably more acceptable the more time that there is before the event, the less meaningful the event, the less it’s putting them in a bind,” he says.

But recent research suggests these social consequences may be overblown. In an unpublished study, researchers found cancelers tend to overestimate how upset others are when they back out and underestimate how much canceling impacted others. “Realizing that canceled-on persons are more understanding than expected can reduce social stress,” the authors write. 

However, this isn’t permission to back out of events with abandon. If you do leave people in a tough spot — they need to cover your concert ticket, they’re scrambling to find another way to get to an event you said you’d drive to — be prepared for scorned parties to share some choice words about you behind your back. “There’s this social penalty, too, that sometimes if you bail, people aren’t crazy about that,” Givi says.

Being in a relationship with someone — friends, family, romantic partners — requires us to honor our commitments. To be there. If we fail to uphold that simple promise, the invites will inevitably dry up. Who wants to keep making plans with someone who seemingly doesn’t value them? How can a friendship blossom when you don’t see one another? 

“If you want a friendship, you have to honor those commitments and show up,” Chopik says. “You have to honor the non-important stuff.”

When I first heard about the new Apple Upgrade program, which lets you lease devices like iPhones and MacBooks for a monthly fee, I was offended. It amounts to paying a tithe to one of the world’s richest companies just to borrow devices for a couple years, rather than buying them outright. You could then choose to purchase the device, which is outdated at that point, or upgrade and keep paying that monthly fee. You may never own an iPhone again.

Then, as my mind wandered to the stack of old phones in my closet, it occurred to me: What’s so great about owning these things to begin with? 

Apple, of course, would love to sell you a new iPhone for keeps. Its most advanced model, the iPhone 17 Pro Max, will set you back $1,200, a price that’s expected to rise soon due to the global shortage of storage and memory chips. You can sign up for an installment plan — most carriers offer these, as does Apple through its credit card — and pay it off in two to three years. Or you could lease the thing for $35 a month under the new Apple Upgrade program. You can pick a 12-, 24-, or 36-month lease, depending on the device, and you don’t get to keep the phone at the end of the term unless you decide to buy it by paying off the remainder of the retail price in one lump sum. (This is similar to the controversial rent-to-own model you find at places like Rent-a-Center.) 

For the financial side of the new program, Apple has partnered with none other than Klarna, the “buy now, pay later” giant. When you go to lease a new device, Klarna runs a soft credit check and decides if you’ll be able to cover the monthly payments. When I asked Klarna, the company did not tell me where it draws the line here, but it’s worth noting that critics have accused Klarna of a lack of underwriting and of lending to people with subprime credit scores. If you miss three consecutive payments, Klarna will terminate the lease agreement and possibly send a collection agency after you.

“How do I know people aren’t getting a good deal here? If they were, Apple wouldn’t be offering it.”

Aaron Perzanowski, University of Michigan law professor

While there was some speculation last week that Apple might lock people out of leased devices if they failed to pay their bill, Apple confirmed to me that it will not put limitations on device functionality due to missed payments or default. If you want to cancel the lease, you face an early termination fee. If you choose to keep paying the monthly fee, you can keep upgrading with new lease agreements for new devices every few years, existing in this cycle indefinitely.

“I don’t think people are getting a good deal here,” said Aaron Perzanowski, a law professor at the University of Michigan and author of The End of Ownership: Personal Property in the Digital Economy. “How do I know people aren’t getting a good deal here? If they were, Apple wouldn’t be offering it.” 

Buy an iPhone? In this economy? 

If you’re someone who likes to get a new iPhone or MacBook on a regular basis, Apple’s new leasing option might make a lot of sense. The monthly fee to lease these devices is cheaper than the payment plan to buy them, and electronics are depreciating assets. If you own one, you can sell it or trade it in for credit toward a new device, but they’re all worth less and less as time goes on. Furthermore, Apple eventually stops supporting old devices through software updates, so they might just stop working at a certain point. Put another way: You may own the phone, but you’re still just licensing the software that makes it work.

Renting an iPhone does sound bleak, though. The United States is suffering through an affordability crisis as prices across the board rise in the face of new tariffs and new wars. Meanwhile, AI is promising to transform the way we work if it doesn’t simply steal our jobs first, adding further insecurity, and the data center boom is making electronics more expensive. This era of economic anxiety is pushing people to use “buy now, pay later” services like Klarna and Affirm to pay for groceries or a tank of gas. (These companies faced scrutiny by state attorneys general a few years ago for operating like predatory lenders.) And now Apple, surely suspecting that many people can’t afford to pay full price for new phones, is inviting us to rent our devices at a monthly fee that undercuts the path to ownership. 

Apple could have just called this the Apple Rental program, by the way. Lease sounds nicer, though, like something you do with a car. 

“It is funny that they frame it as not a loan but as a lease,” Louis Hyman, a history professor at Johns Hopkins University and author of Debtor Nation: The History of America in Red Ink. He added that “leasing” has class implications, suggesting that you’re either someone who needs to have the newest things but can’t afford them, or that you’re so wealthy, you’re indifferent to money.

Suffice it to say, the bulk of people who will soon be leasing their iPhones are probably not the ones who are indifferent to money.

Apple adopts its final form

The new Apple Upgrade program is the company’s latest customer acquisition strategy. As the rising price of hardware has made cheaper Android devices or the refurbished market more attractive, Apple is offering upgrade enthusiasts and budget-minded users, including people who simply couldn’t afford to buy Apple products in the past, a deal to join the company’s ecosystem. After all, keeping people supplied with new iPhones and MacBooks also helps keep them subscribed to Apple services, like iCloud, which now makes the company more money than Mac, iPad, Apple Watch, and other accessories combined.

If Apple’s financial future hinges on getting more and more people to subscribe to these services, it’s only natural that the company would want to lower the barrier to entry. So Apple is betting that by letting people use but not own its products, it will extract more profit in the long run through lease payments and subscription fees. After all, it wasn’t that long ago that it seemed like nobody was interested in upgrading their iPhone, since the new phones looked so much like the old ones. Now, Apple is just trying to get everyone on autopay, effectively subscribing so that they get the latest devices when they come out.

There’s not necessarily any harm in giving people a cheaper way to access expensive but useful products. For more than a century, installment plans have enabled people to buy modern conveniences like sewing machines, radios, and eventually, televisions. Leasing is a popular way to keep yourself in a new car, sometimes with free maintenance. Meanwhile, cellular carriers have a long history of helping their customers buy phones. Nearly two decades ago, you could get an iPhone 3G for $199, thanks to subsidies from AT&T, which the company recouped in service fees over the course of your contract. Sprint and T-Mobile have even offered unlimited upgrades through leasing programs of their own in years past.

Apple previously worked with Citizen One Bank to offer loans to customers who wanted the option to upgrade their iPhones every year. The payments were higher and they included a fee for AppleCare, but every year, you could trade in your current phone for a new one. If you didn’t want to upgrade, you could simply keep paying the installments, and you’d eventually own the phone. Most carriers now give you the option to set up a payment plan to purchase a new device that simply amounts to the retail price of the gadget divided by the number of months you’ll need to pay it off, usually 24 or 36, with zero interest. That makes it easier to get your hands on an iPhone Pro Max, and if you pay it off in full, it’s yours for life — or until Apple convinces you to buy another new iPhone.

The difference between paying those monthly installments and paying a monthly lease agreement, of course, is that the former puts you on the path to ownership. The latter simply puts you on a path to make a decision: Do you want to buy the thing and recoup some of the money you’ve already spent, or do you want to keep making payments?  

“What ownership ideally gets us is independence,” Perzanowski said. “It gives us autonomy. It gives us the ability to function in the world without relying on third parties.” He went on to explain how moving from owning a product to leasing it means you’re stuck with that third party. “I’m tied to that manufacturer in a way where they get to exert a fair amount of control over my behavior,” Perzanowski said. “Historically, we’ve been primed, especially in the United States, to resist and reject that kind of control.”

One great thing about owning an iPhone or a MacBook outright is that if you lose your job to AI, you don’t have to come up with a monthly payment in order to keep using those devices to apply for new jobs. Another great thing about ownership is that should you need a couple hundred bucks, you can sell that old phone or laptop and pocket the cash. Maybe the best thing about owning these devices is that you can repair them and keep using them for many years — or at least until Apple stops supporting them. 

That doesn’t mean leasing never makes sense. If your digital life revolves around always having the newest devices and you upgrade every year or two no matter what, you might actually save money by doing so through Apple’s leasing program. If you need an iPhone or MacBook right away but can’t afford to pay full price or even cover the monthly payments on an installment plan, a one-year lease could be a good solution. 

Invariably, when you lease anything, you’re entering into a contract, one that comes with consequences if you break it. Leasing an iPhone means you’re tied not only to Apple but also to Klarna for the next 12 to 36 months. If something goes wrong — you lose your job, you lose or break your phone, or you simply don’t want the device any more — you’re subject to the terms and conditions of these big tech companies. If you keep renewing your lease, you may very well end up spending more on a phone than you would have if you’d bought it outright. That would be fine with Apple, of course. It has shareholders to please.

Correction, July 30, 1 pm: This story originally misstated how the previous Apple upgrade loan program worked; it allowed phone trade-ins every year, not every two years. 

The AI industry’s investors and critics don’t agree on much. But many in each camp share at least one basic conviction: America’s top labs are about to make a killing. 

Capital markets have signaled their faith in Anthropic and OpenAI’s impending hyper-profitability, valuing each at nearly $1 trillion. Many of Silicon Valley’s progressive adversaries also expect the labs to grow filthy rich but fear the implications, warning that AI-induced automation could transfer vast sums of money from ordinary workers to a handful of giant tech companies. Sen. Bernie Sanders’s call for nationalizing the top AI labs rests partly on that concern. 

But recent advances in Chinese AI call all of this into question.

Over the past two months, Chinese companies have released three AI models that are nearly as powerful as America’s frontier systems — and radically less expensive. 

In June, Beijing’s Z.ai debuted a model that performed nearly as well as Claude and ChatGPT’s second-tier systems on independent benchmarks. Weeks later, another Chinese firm, Moonshot, unveiled “Kimi K3,” a model that allegedly outperforms all of its American rivals except for the very latest versions of Claude and ChatGPT. Finally, just days ago, Alibaba launched a preview of Qwen3.8 Max, which purportedly outclasses even OpenAI’s most advanced systems, while trailing only Claude’s Fable in its capabilities. (Disclosure: Vox Media is one of several publishers that have signed partnership agreements with OpenAI. Our reporting remains editorially independent.)

These developments don’t merely threaten America’s AI giants with stiffer competition in the race for superintelligence. Rather, they raise a more harrowing prospect: that the AI race’s ultimate rewards will be far smaller than anticipated. In a world where new advances can regularly be leapfrogged by cheaper upstarts, hoarding the technology — and its profits — will be harder for any one company to do.

In other words, building a machine God might not be as lucrative as it’s cracked up to be. AI, it turns out, may “want to be free.”

How AI was supposed to pay off

To see how China’s new models threaten Anthropic’s profit expectations, we must first examine why those expectations have been so high.

This is not entirely self-evident. After all, AI labs aren’t much like the hyper-profitable tech giants of the 2010s. Facebook and Airbrb were relatively capital-light businesses with ultra-low marginal costs (adding a profile to Facebook or listing to Airbnb costs the companies virtually nothing). And once each gained a foothold in their respective markets, network effects enabled them to retain formidable positions without needing to constantly upgrade their products.

Building a state-of-the-art AI company is a much more involved — and astronomically more expensive — endeavor. To get to the frontier, Anthropic and OpenAI have sunk (at least) tens of billions into semiconductors, data centers, power plants, and other capital investments. Staying at the cutting-edge, meanwhile, compels them to perpetually churn out evermore costly models.

To put a new Claude model through its initial training — in which it spends months digesting the internet and sussing out statistical patterns within its text — can now cost hundreds of millions of dollars. And such foundational computation is only the beginning. A truly superlative model requires several additional months of fine-tuning. Armies of contracted experts — such as computer scientists, physicians, and mathematicians — tutor the models, grading their answers and guiding them towards better ones. Then the AI systems complete millions of rounds of practice, in which they learn through trial and error how to solve countless problems. This arduous process, known as “post-training,” compounds the costs of a single model’s development. 

All of which raises the question: Why would investors expect businesses with a cost-structure this challenging to be not merely profitable, but massively so?

There are (at least) two answers. The first (and most obvious) is that the market for superintelligent machines is liable to be vast. Frontier AI systems promise to reduce costs and improve performance in myriad white-collar sectors. And Anthropic’s soaring revenues indicate that firms do, in fact, find Claude useful. A company like AirBnB has earned billions by revolutionizing a single industry; imagine then what a technology that remade virtually all industries might be worth.

Of course, plenty of technologies are valuable but not massively profitable to produce. After all, in well-functioning markets, competition should eventually erode individual firms’ margins, even if the underlying technology continues generating huge value. 

But this is where the second answer comes in: Frontier labs’ immense costs are a burden, but they’re also a safeguard against competition — or, in industry parlance, a “moat.”

Startups may be able to afford to build or acquire more rudimentary models, many of which are “open source.” But, the thinking goes, they won’t be able to deliver Claude Fable-level performance without raising giant amounts of capital. And what investors will be willing to pour hundreds of billions into an AI pipsqueak that’s light-years behind Google, Anthropic, and OpenAI?

Alas, the Chinese AI labs’ rapid progress — and the way it was achieved — suggest that Anthropic’s moat may be shallower than previously thought.

How Moonshot swam Anthropic’s moat

The existence of powerful, Chinese AI systems is neither new nor surprising. Xi Jinping’s government has made vying for global AI dominance a key economic goal. And China’s DeepSeek, which also has stunned US companies with its lower-cost competitive models, surpassed ChatGPT as the most-downloaded free iPhone app more than a year ago.

The latest models, however, have dramatically narrowed the gap in capabilities between frontier American systems and their Chinese rivals. Just as critically, they’ve done so in a manner that other, relatively underfunded AI upstarts might be able to emulate.

Alibaba and Moonshot needed to invest massive resources to train their base models. But they allegedly found a low-cost way to refine those models into near-frontier systems: Just ask Claude.

Or, more specifically: Engage Claude in 16 million conversations, using 24,000 fake accounts. In each of those exchanges, ask the model to not only answer countless difficult questions but also, walk you through its reasoning, step by step. Then take all of this data and feed it into your own model as study material, training it to respond to the world’s most challenging queries as Claude would. 

Through this process — known as “distillation” — an AI lab can replicate virtually all of a frontier model’s capacities, without sinking vast sums into human experts and post-training computing runs. 

China’s AI labs have not admitted to using distillation. But OpenAI and Anthropic both reportedly uncovered Chinese distillation attempts earlier this year. And some of the new models appear to display tell-tale signs of distillation in conversations with ordinary users; Kimi K3 has routinely identified itself as “Claude.”

Chinese AI companies are hardly alone in using distillation to catch up with frontier labs. Earlier this year, Elon Musk admitted in court that xAI enhanced Grok’s capabilities by running distillation techniques on Claude and ChatGPT. Nonetheless, China’s latest models appear to demonstrate that distillation can help take a second-tier model to the frontier’s threshold.

America’s frontier labs have tried to defend themselves against such imitators. But this is technically difficult when distillers can assemble massive networks of bots, each asking an inconspicuous number of questions. And legally, it is difficult for America’s AI giants to argue that distillers are stealing their intellectual property. After all, in a sense, China’s copycats are merely doing to Anthropic and OpenAI what those companies did to journalists, coders, lawyers and other specialists: Feeding their public-facing outputs into a model, which then replicates their capabilities by discerning underlying patterns within the text.

Oh, and China’s giving these models away

The new Chinese models would have caused Silicon Valley enough headaches, if they merely provided stiffer competition, while demonstrating the power of distillation. 

What makes Kimi K3 and Qwen3.8 Max especially threatening to the American AI giants’ profitmaking potential, however, is that they are officially open source — meaning that the models’ parameters can be downloaded for free. (Alibaba and Moonshot have not yet released these parameters, but they say they will shortly.)

In other words, any company or hobbyist with enough computing power will soon be able to run a near-frontier Chinese model on their own hardware, modify that model to better serve a specialized purpose, and then sell access to their new version — without paying Alibiba a single yuan.

As Kimi and Qwen grow more capable, their market-share is likely to grow, at American AI giants’ expense.

For many of Anthropic and OpenAI’s potential customers, that proposition may be hard to turn down. Most businesses don’t need the world’s smartest AI, just one competent at their enterprise’s core tasks — compiling legal research, answering IT queries, writing working code, etc. A model that produces outputs 90 percent as good as Claude’s — at roughly one-sixth of the cost — will sound pretty good to many corporations.

Further, open source models aren’t just cheaper than frontier systems, but potentially more secure. If you run an AI on your firm’s own servers, then you don’t need to entrust sensitive data to Anthropic, Google, or OpenAI.

All this had led much of corporate America to embrace open-source models, even before the latest versions narrowed the capabilities gap. In a Linux Foundation survey, 63 percent of organizations reported using open-source AI systems.

And increasingly, those models are Chinese. According to Sequoia Capital, one of Silicon Valley’s premier venture capitalist firms, a majority of American AI startups now use open-source Chinese systems. As Kimi and Qwen grow more capable, their market-share is likely to grow, at American AI giants’ expense.

What’s bad for OpenAI is good (and/or catastrophic) for humanity

All this said, it is still entirely possible that OpenAI and Anthropic will justify their colossal valuations. In many highly competitive economic domains, having access to the world’s very best AI model will remain highly valuable. And America’s frontier labs still outperform all their peers. 

But it’s increasingly plausible that selling state-of-the-art AI systems will prove to be a low-margin undertaking. In a world of ubiquitous, near-frontier open source models, the AI sector’s big winners probably won’t be its top labs, but rather, its chipmakers and cloud computing providers. 

For ordinary people, a future where superintelligence is dirt cheap — and rival AI companies are constantly rising and falling, rather than consolidating into mega-corporations — would look somewhat different than the cyberpunk dystopia that the left’s been dreading. 

And not entirely in a good way. For one thing, in that reality, mitigating AI’s biggest risks would be immensely difficult. Having a handful of firms monopolize control over frontier AI systems is bad in many respects. But it does make those models easier to regulate, as the Trump administration’s decision to temporarily block Claude’s Fable in the name of cybersecurity demonstrated. 

By contrast, if recipes for ultra-powerful AI models are published all over the internet — and anyone with modest technical skills can modify them at will — then systems willing to help their users hack government bureaucracies or engineer bio-weapons are liable to proliferate.

From another angle, however, the “AI becomes almost free” scenario may look like capitalism at its finest: Retrospectively, such a development would mean that a small number of extremely rich people bankrolled the creation of an immensely useful technology, under the expectation of massive profits, only to see competition erode their returns — and disperse that tech’s benefits across a wider group of businesses and consumers. 

Granted, in the case of AI, this process might also generate a super-virus that kills us all. But hey, no system is perfect.

This story appeared in Today, Explained, a daily newsletter that helps you understand the most compelling news and stories of the day. Subscribe here.

Surely Anthony Fauci has earned the right to party with Robert De Niro in peace. But instead, America’s embattled former Covid-19 czar was hauled before the Senate today to defend five-year-old diary entries. 

Senate Republicans subpoenaed Fauci, the former director of the National Institute of Allergy and Infectious Diseases, as part of a broader effort to dispute the origins of the Covid-19 pandemic.

Fauci is the longtime face of American infectious disease research and has appeared in front of Congress more than 250 times. But this morning — in a tense, highly partisan spectacle — he invoked the Fifth Amendment and refused to testify. 

Fauci accused Kentucky Sen. Rand Paul, who’s leading this circus, of nursing an “unhinged” and “obvious obsession” with him. Paul recently released Fauci’s unredacted diary, which includes lots of deeply cringe tidbits about the immunologist’s celebrity hobnobbing…but nothing that materially alters our understanding of the pandemic or Fauci’s handling of it. 

“The only conclusion I can reach is that the sole reason he is calling me before this committee is to get me to say something, anything, that could vindicate his repeated public pledges that I end up, in his words, ‘behind bars,’” Fauci said. “Any reasonable person who has followed his unhinged obsession with me would readily come to the same conclusion.”

Why we’re still arguing about Covid-19

Fauci occupies an uneasy spot in American politics. To some people, he’s a hero who helped steer the country through a terrifying pandemic. To others, he’s a villain who imposed damaging policies, concealed inconvenient facts, and never answered for his various missteps.

The evidence doesn’t back that darker story. But Fauci’s vilification points to one possibility that is worth taking seriously: As a country, we never adequately reckoned with the mistakes and uncertainties of Covid-19.

Consider the narrative trajectory of the lab-leak theory, which is central to Paul’s vendetta against Fauci. Early in the pandemic, Paul and other proponents argued that the novel coronavirus might have escaped from a Chinese laboratory. Some people went even further, claiming that Fauci had funded the lab’s research and then covered it up. Both theories were treated as ridiculous.

To be clear, the second theory is ridiculous. (There’s absolutely no evidence that Fauci was somehow, improbably involved in the making of Covid-19.) But the consensus around the lab-leak theory has changed. While most researchers still believe the virus spread naturally from animals to humans, a laboratory accident has never been definitively ruled out — and many now allow for the possibility in a way that they didn’t four or five years ago. 

The same is true of other pandemic controversies, too. Reasonable people can now debate whether lengthy school closures did more harm than good. Guidance on masks and other measures changed repeatedly, sometimes without a ton of explanation. 

None of that means officials acted in bad faith. But we’ve arguably never had the kind of serious, bipartisan examination of the pandemic’s missteps that a once-in-a-century public health crisis deserved. And in the absence of that (naive? impossible?) shared reckoning…we’re getting Fauci’s leaked journals. 

One link for later

➨ Check for ticks. Ticks are responsible for a whopping 90 percent of vector-borne disease in the US, but local and county health departments do very little to stop them. To protect yourself, use permethrin spray when you go outside and examine your body for the little buggers when you get home. Throw your clothes in the dryer after a hike, too — ticks hate heat! Who knew. 

Before you go…

• Did you know…that in Wisconsin, placing certain bets on the prediction market Kalshi can cost you your right to vote? The law dates back to 1849, and it’s one of several state laws that bar voters from gambling on elections. 
• Today’s trivia: In Greek mythology, who is Hercules’s stepmother? (You can find this and other brain puzzles in Vox’s daily crossword. Look for the answer in tomorrow’s edition.)
• Yesterday’s trivia: Yesterday we asked you the name of the Dutch cheese typically coated in red wax. That would be the Edam. In the 15th century, Edam reportedly helped make its namesake city one of Europe’s most important cheese markets.