Three hikers rescued from Mount Shasta said they'd relied too much on AI, the Forest Service said.
Siskiyou County Sheriff's Office
Three hikers rescued from California's Mount Shasta on Monday said they "relied too much" on AI.
Ranger Nick Meyers said it was the third time this year that AI played a role in hiker mishaps.
One woman, he said, ignored a "no camping" sign because AI told her it was a good place to camp.
Nick Meyers, lead climbing ranger for the US Forest Service on Mount Shasta in California, has seen people make plenty of mistakes in the mountains over the past 20-plus years. But he's seeing a new trend in 2026: hikers erring with AI's help.
Meyers, who is also the director of the Mount Shasta Avalanche Center, said there have been at least three mishaps this year in which hikers or campers were led astray by AI or relied too heavily on the tech.
The most consequential came last weekend, when three 20-year-old men got lost and had to be rescued after summiting Mount Shasta, California's fifth-highest peak, located in the Cascade Range at the northern end of the state.
The "novice climbers" had set out on Saturday on the Clear Creek route and established base camp at a prohibited site, the US Forest Service said. They woke up at 3 a.m. on Sunday and set out for the summit, equipped with enough food and water for an eight-hour hike. They summited 16 hours later at 7 p.m.
While trying to descend the 14,179-foot peak in the dark, the climbers went off route, and one of them injured his knee. The cellphone they'd been using for navigation died. They eventually had to stop and form a makeshift camp to spend the night, without adequate food, water, or equipment, the Forest Service said. At 9:30 a.m. on Monday, climbing rangers reached the party and helped them descend.
The group told the rangers they had used Google Gemini, YouTube, and the AllTrails app to plan their climb, authorities said.
"We relied too much on AI rather than our own critical thinking," the group told rangers, the Forest Service said.
Meyers said AI was not the "nail in the coffin" for the group, but that they had made a series of mistakes that all added up. He said the roughly eight-hour estimate they got from AI wasn't generally wrong, but that the group hiked far slower than that, hadn't packed enough provisions for the extra time, and kept going after blowing past the recommended turnaround time of around noon.
Still, he said the group itself believed AI contributed to their mistakes and that there may be too much reliance on the technology, especially among younger hikers.
Google did not respond to a request for comment. The company says on its website that Gemini can make mistakes and hallucinate, and that users should double-check its responses.
The hikers rescued on Mount Shasta said they used Google Gemini to plan their trip.
U.S. Forest Service, Shasta-Trinity National Forest
AI-related mishaps are on the rise
Meyers said rangers encountered two other people this summer who ran into issues after using AI to plan their trip.
He said one woman had set up camp in an area where camping was not allowed, as was clearly labeled at the site. When rangers informed her she couldn't camp there, she said, according to Meyers, "Well, ChatGPT told me I could camp here."
"We're like, 'Well, the sign says no camping,'" Meyers said.
In another incident, a man was talking to rangers about a hiking loop he was planning to take, but they were confused by what he was saying. When they asked exactly what he was planning, he pulled out a AI-generated map of the mountain and his route.
"The map was terrible, totally terrible," Meyers said. "It had features that were on complete wrong sides of the mountain."
OpenAI says on its website that ChatGPT can make mistakes and might respond confidently "even when it's wrong." The company also lists practical tips for how to use its chatbot responsibly, including: "Use ChatGPT as a first draft, not a final source. Please check important information."
People making mistakes in the mountains is nothing new, and rangers have long advised hikers to do their homework, plan carefully, and consult official sources, such as the Forest Service or the National Park Service. AI can be used as one part of that planning, too.
"AI can be a great tool," Meyers said. "But it's not the only tool in the toolbox."
AI is especially dangerous to rely on in the mountains, where you're dealing with Mother Nature, unpredictable weather, and risky terrain. Meyers said hikers also need to rely on what he called "old school skills" — like talking to a real person, a ranger or an experienced climber, and getting information from firsthand sources.
Mount Shasta specifically has a website updated with the latest information, and climbers are encouraged to call to speak with a ranger and ask questions. AI can also be used to find the real people you need to call.
"Our website will pop up," he said. "We've queried it."
The reality is that safety in the mountains is dependent on a series of small decisions you make over and over again as you're climbing, Meyers said, and part of that is based on "mountain sense" built up through experience.
"One thing we don't ever want to do is discourage people from adventures," Meyers said. "We just want to make sure they plan properly and encourage folks to start at the shallow end of the pool."
His advice: Find a good mentor. Talk to real people. Visit the ranger station.
"You can do your online research," he said, but also "fact-check yourself a little bit and use those critical thinking skills."
Do you have a story to share about using AI in the outdoors? Contact this reporter at kvlamis@businessinsider.com.
Ilana Glazer speaking into a megaphone at an anti-AI rally in New York City in June.
Courtesy/Greg Pak
Mamdani paused most AI use in schools this week for at least a year.
The decision came after parents on both sides of the issue lobbied his government for months.
One side argued that AI is bad for a child's development. The other: It's necessary for a career.
Craig Garrett had no idea his 5-year-old daughter was talking to what he described as an AI chatbot at school every day.
That was until the Brooklyn-based book editor attended a meeting held by his school district, where he said educators revealed the extent of AI use in his daughter's classroom.
"I was kind of horrified and shocked and disappointed," Garrett told Business Insider. "The whole notion of a 5-year-old plugged into a tablet with their headphones on and a microphone, and speaking to this animated character that's talking to them like it's their friend is just deeply wrong."
That meeting, which took place in May last year, "radicalized" many of the parents in the room, Garrett said. It led him to join the AIM coalition, a group of New York City parents opposed to AI in the classroom. AIM stands for AI Moratorium.
Those concerned parents are often the loudest voices in the room, but they aren't the only concerned parents, and they aren't the only voices. There is a cohort on the other side that believes students preparing to enter the workforce need to be taught to use AI, or else be left behind.
"In the last six months alone, the references to having AI skills in job postings in the United States have more than doubled," Keri Rodrigues, the founding president of the National Parents Union, which has a chapter in New York City and which supports AI use in schools, told Business Insider. "Parents want their kids to understand how to utilize this technology."
These two groups of parents, representing opposite sides of the debate, engaged in a monthslong effort to lobby Mayor Zohran Mamdani's administration, which this week ultimately sided — at least for now— with the anti-AI contingent.
Mamdani announced on Wednesday that his administration would impose a one-year moratorium on student-facing AI platforms in New York City public schools for students in elementary and middle school. Under the new restrictions, Amira, the program Garrett's daughter was using, will have its AI features turned off.
A spokesperson for New York City Public Schools said in a statement that the school system "engaged with various stakeholders — educators, students, parents, advocates, union partners, experts, and other community members" to shape the policy.
The AIM coalition had two meetings with NYCPS Chancellor Kevin Samuels in recent months, members said.
Parents in the AIM coalition called the moratorium a victory, though it fell short of their demand to pause the tech for two years.They described feeling blindsided by the extent of AI use in the classroom. Some said they feared that data about their children would be scraped by AI companies. Others said they didn't want chatbots taking the place of teachers.
"It really started from concerned parents finding out things about what their children were experiencing and being completely appalled," said Liat Olenick, a leader of the group. "It really took off because I think it struck a nerve."
Olenick recalled multiple incidents of parents approaching her in bewilderment about the AI programs their children were using. "All of a sudden, one day, people were like, 'Wait, my 7-year-old's using AI? I never gave permission for that,'" she said.
Olenick said the group began campaigning for a moratorium earlier this year. She said the group's lobbying efforts included conversations with staff members in the mayor's office, as well as the meetings with Samuels.
"The organizing really made a difference," she said. "We changed the conversation."
Members of the AIM coalition pictured after a meeting with New York City Public Schools Chancellor Kevin Samuels at the Tweed Courthouse in May.
Courtesy/Craig Garrett
Under the new AI guidelines unveiled on Wednesday, teachers will still have access to generative AI tools, which they can use to plan lessons but not to grade assignments. High school students will participate in "AI literacy" courses and have limited access to five pilot AI programs, none of which are chatbots.
Joe Philleo, the CEO of Edia, one of the programs permitted for high school use, applauded the one-year pause in a post on X. "Mamdani's plan strikes a reasonable balance for this school year," he wrote. "Young kids should not be glued to their screens all day."
The CEO of Amira, meanwhile, defended the software's role as a tool for teachers.
"We understand how challenging it is for districts to ensure that they're putting appropriate AI safeguards in place, and appreciate that, as New York does that hard work, they're also taking care to ensure that teachers who are using Amira to diagnose reading challenges and improve literacy outcomes for kids can still access the tools that are working for them," the company's CEO, Mark Angel, said in a statement.
Mamdani described the moratorium as the most "comprehensive of its kind anywhere in the country." It comes during a nationwide public reckoning over how best to expose children to generative AI, as educators bemoan how students are using the technology to cheat on assignments and parents demand more regulation.
Rodrigues, however, said the ban was a "hysterical response"that could leave students unprepared for a job market that values AI skills more than ever.
Though Rodrigues said AI has its place in the classroom, particularly when combined with strong and consistent oversight by human teachers, she stressed that there are certain functions — like grading, counseling, and imposing discipline — where AI has no place.
The new policy, she said, uses a sledgehammer instead of a scalpel.
"You are seeing a knee-jerk political response to something that is complex, requires nuance, requires foundational understanding, and should be evidence and data-driven," she said.
City officials emphasized during their press conference announcing the moratorium on Wednesday that the pause was temporary as they assess how AI affects a child's ability to learn.
Kaliris Salas-Ramirez, a neuroscientist and associate professor at Rutgers University, and a mother of two sons in New York City's public school system who supports the pause, said research suggests young children using AI tools risk preventing their brains from developing problem-solving skills.
"We're actually not allowing the brain to consolidate that information in a comprehensive way," she said, referring to the use of AI during a young person's formative years.
For Garrett, whose daughter is now entering second grade, he's relieved the Amira app won't be used as often.
"It's not like this app was doing something that teachers couldn't do before," he said.
US Energy Secretary Chris Wright oversaw the signing of a series of energy deals during his visit to Caracas, where he met interim President Delcy Rodríguez.
Juan Barreto/AFP/Getty Images
The US could swap Venezuela's heavy crude for lighter American oil to help refill the SPR.
The Venezuelan barrels don't have to enter the reserve themselves, Energy Secretary Chris Wright said.
More Venezuelan heavy crude could also mean more competition for Canadian oil.
The US could swap heavy Venezuelan crude for American oil to help refill the US emergency reserve, Energy Secretary Chris Wright said Wednesday.
"We'll swap you a barrel of heavy crude for a barrel of light or medium United States crude," Wright told CNBC in Caracas.
The swap could get around a mismatch, as some of Venezuela's key heavy crude grades aren't suitable for the Strategic Petroleum Reserve.
Wright's comments came after President Donald Trump said Sunday that he planned to "fill up" the US strategic reserves with Venezuelan oil. The White House has also said oil secured through a sweeping new US-Venezuela agreement could help replenish the SPR.
"It doesn't mean that exactly these barrels go in there," Wright said.
Venezuela's heavy crude could instead go to US refiners, while lighter American barrels obtained through a swap could enter the reserve. Wright said US refineries built in the 1960s and 1970s were designed around Venezuelan crude.
Under the new agreement, North American Blue Energy Partners received 100-year concessions covering 17 Venezuelan oil fields. The US government received a 35% equity stake in the company's corporate parent. The State Department gained the right to buy 20% of the oil from NABEP-operated fields at the cost of production.
"Well, of course. All crude is competition with all other crude," Wright said when asked whether Venezuelan production would compete with Canadian and other North American oil.
Wright said increased competition would push prices lower, which he expects to boost consumption.
"If you have lower oil prices, you have faster demand growth," he said.
Domino's is finally launching a truly personal pizza. Yay! Right? Except that the company's announcement about it is weirdly sad. Hungry diners will be able to customize "without compromise," it says, avoid "sacrifice," and share a meal without actually sharing. Of course, pizza's not that deep, but that's sort of the point. We should generally be able to come to an agreement on olives and pepperoni.
A growing number of Americans are eating by themselves — the Bureau of Labor Statistics' American Time Use Survey shows that as of 2023, over a quarter of Americans spent most of their eating and drinking time alone, up from a fifth in 2008. When people do end up gathering for meals, they're increasingly siloed. We've figured out how to eat alone even when we are together.
Food hall-inspired startup Wonder makes meals from multiple menus out of the same kitchen, so no one has to choose between cuisines. Delivery company DoorDash has launched a feature for business customers that lets employees choose separate lunches from different places and get them delivered all at once. A 2026 OpenTable survey found that 52% of Americans say they'd rather order their own dish than share with the table, even as group dining overall is up.
We may be sharing space, but we're not sharing food — and, in turn, we're not sharing all the benefits that come with a communal plate.
"The act of coordinating itself makes people have trust and connection," says Kaitlin Woolley, a professor of management at Cornell University. "Sometimes, this little bit of friction or this little challenge, it does help us to come together."
Woolley means what she says quite literally. In one of her studies, she paired people off and had them either share a bowl of chips and salsa or each get their own. She then put them in a negotiation scenario, in which one person played management and the other a union representative, and asked them to reach a wage agreement. Her findings: Pairs who shared a snack struck a deal faster than those who didn't. The slight inconvenience of timing chip dipping and trying not to hog helped the partners forge an understanding.
Sharing a meal often involves a series of mundane accommodations — choosing the restaurant, waiting until everyone has food before starting, not eating too fast, leaving the last cookie on the tray until someone finally relents and takes it. None of this behavior is particularly profound, but it makes us pay a little more attention to others' wants and actions.
When people order from the same venue or eat the same cuisine, the collective experience improves. That comes at the small expense of everyone not getting precisely what they would have chosen were they on their own, says Ayelet Fishbach, a professor of behavioral science and marketing at the University of Chicago. "There's a lot to gain from the social side, and the individual cost is usually negligible," she says.
Sharing a meal often involves a series of mundane accommodations.
We tend to bond over similarities related to food as well, Woolley adds: "More so than taste in music or similar clothes, it seemed to be a way for people to connect."
This isn't always as simple as everyone agreeing on the same cheese pizza and moving on. Many people have food allergies or dietary restrictions, and the option to personalize things makes it easier to participate, even if it comes with its own set of challenges. Instead of getting veto power, the person with the issue can do their own thing and still hang out.
It can be stressful to be the odd one out, and research shows that people with food restrictions tend to worry about how they'll be perceived. They get anxious about having to reveal their restrictions, wonder whether people will decline to invite them in the future, and feel self-conscious about others making negative assumptions about them.
Kevin Kniffin, an assistant professor of management and organizations at Cornell University, did a study about a decade ago that found that firefighters who eat meals together perform better as a group than teams who eat alone. One of the firefighters in the study was vegan. He would bring his own meal, but still made sure to eat at the same time and at the same table as everyone else. He also took part in shared clean-up chores, despite not contributing to the shared mess.
"The firefighter seemingly had a strong intuition that all of those shoulder-to-shoulder and face-to-face activities were important, and it was his smart way of matching his own dietary decisions with his platoon's meal-sharing practices," Kniffin says. He adjusted the part of the meal he needed to without opting out of everything around it.
To be sure, people share meals without eating the same food all the time — i.e., we often order separate dishes at restaurants. Eating distinct items together doesn't erase social interaction. A lot of the dynamics depend on expectations. If everyone agrees to bring their own lunch to the breakroom, that can strengthen bonds. But if someone invites friends over with the express purpose of cooking a homemade meal and a guest decides to DoorDash their own food instead, that's a no-no, even if the host might not openly show offense.
"There are silent social agreements," says Nicklas Neuman, an associate professor and senior lecturer in food studies, nutrition, and dietetics at Uppsala University.
The idea that everyone should get exactly what they want is a modern luxury. It's a "radical" development in human history to choose what we eat based on taste preference rather than what's available to us, Neuman says. "It's obviously also connected to how individualized the culture is," he says.
Meals aren't just about nutrition, taste, or social gatherings — they're also a consumer choice. What we eat says something about who we are, what we value, what we can afford, and where we are in the class hierarchy.
Delivery apps and the ability to get hyper-customized meals have unbundled the constellation of decisions and activities typically involved in communal eating, Kniffin says. In terms of the consequences, he compares the effects to Jenga: It's not clear "which pieces in the bundle are critical" to actually providing the societal benefits of a shared meal. Maybe the shared entrée isn't load-bearing, but if you remove too many other blocks, the whole thing collapses, and what remains is a handful of people consuming individualized products around the same time, staring at their phones.
Apart from food selection, other parts of the shared meal are turning into atomized interactions as well. Technology has enabled the precise division of the bill. A multi-person lunch or dinner has long created opportunities for low-stakes financial tensions. Do we split it evenly? Does the person who skipped drinks pay less? Does anyone really want to decipher who ate how much of the appetizer?
Everyone choosing their own specific adventure may lead to a trade-off between long-term social connection and short-term convenience.
Apps such as Zelle, Splitwise, and Venmo make it possible to settle those questions with strict accuracy. That may ease potential conflicts, says Margaret Clark, a professor of psychology at Yale University. When one friend starts calculating everyone's share down to the last cent, "it just feels petty," she says. If, instead, the app automatically does the arithmetic, it may feel less loaded. "It's not the person, it's not one of the members, who is suggesting that we should be transactional to the last cent," she says.
Friction at mealtime isn't inherently virtuous, nor is removing it necessarily antisocial. Sometimes, technology and optionality clear away annoying logistics and make it easier to be together. Other times, the thing being engineered away has value — taking turns, making accommodations, finding compromise. Everyone choosing their own specific adventure may lead to a trade-off between long-term social connection and short-term convenience. There's something to approaching meals in a more joint, straightforward way — we eat the same thing, generally guess how the money should shake out, or you say you'll get me next time when I pay — that might make the relationship better.
Maybe instead of five people ordering five individual pizzas and going with Domino's "without compromise" approach, most of the group agrees to get a large cheese. The person who really wanted sausage learns to live without it for a meal, and the person with an allergy gets a separate small pie with a gluten-free crust. Everyone tackles their slices around the table and discusses whether it might be better to do Chinese or Mexican next time.
Emily Stewart is a senior correspondent at Business Insider, writing about business and the economy.
According to a new survey, the share of burners in their 20s has plunged to 10% in 2025.
Getty Images
Gen Zers made up just 10% of Burning Man attendees in 2025.
The share of burners in their 20s has plunged from more than 30% a decade earlier.
The cost of attendance was cited in 2025 as the main obstacle to being at the festival.
Burning Man is falling out of favor with Gen Zers.
As the annual counterculture festival kicks off in the Black Rock Desert of northwestern Nevada, the annual Black Rock City census survey, which tracks the demographics of festival attendees, shows that burners in their 20s hit a record low in 2025.
According to the survey, around 10% of surveyed burners in 2025 are in their 20s at the time of attendance, compared to more than 30% back in 2015. The figure has been in steady decline over the past decade, but it took a much steeper nosedive after the festival resumed in 2022 following a two-year pandemic hiatus.
The survey, which also asks respondents about their biggest obstacles to attending the festival, provided insights into why younger attendees feel deterred.
In 2025, more than a quarter of survey respondents said that the cost of attendance was their largest obstacle, while another 16% cited getting time off from work or school. More than 10% also said that learning how to prepare for a week in the desert is a challenge, since participants are expected to bring their own water, food, and camping gear.
The price of admission to Burning Man has climbed over the past decade, and the Burning Man Project, the organization behind the festival, has faced funding shortages in recent years as ticket prices have struggled to offset operational costs. According to the organization's financial disclosures, it lost $265 per main ticket sold in 2024.
Compared to the main ticket price of $390 in 2015, Burning Man tickets started at $550 in 2026, while a vehicle pass cost another $165. That put the lowest advertised price for someone driving alone at $715 before taxes and processing fees. Including camping gear, food, water, transportation, and camping fees, those costs can easily push the total price of attending beyond $1,000, particularly for first-timers.
The Burning Man Project did not immediately respond to a request for comments.
Brett Schreiber, a plaintiff trial lawyer who led the first jury win against Tesla involving Autopilot, said he continues to receive inquiries about potential cases involving the EV maker's Autopilot and Full Self-Driving systems.
Courtesy Brett Schreiber
Brett Schreiber led the first Tesla Autopilot case in which a jury found the company liable.
Schreiber said he has more inquiries about potential cases related to Tesla's Autopilot and FSD.
He said he has to reject many of them because taking Tesla to court is costly.
This as-told-to essay is based on three conversations with Brett Schreiber, a California-based lawyer who represented the plaintiffs in the Miami Autopilot case involving Naibel Benavides Leon. In 2025, a federal jury found Tesla partially responsible for the crash that killed Benavides Leon and seriously injured Dillon Angulo. The jury ordered Tesla to pay more than $242.5 million in damages. Tesla is appealing the verdict. Tesla and its attorney did not respond to requests for comment. The conversations have been edited for length and clarity.
I've practiced law in California since 2005.
I came up in a plaintiff civil litigation firm in San Diego and had the opportunity to work up and try serious-injury cases.
Around 2021, I started getting calls about Tesla crashes. At the time, I had a GED-level education in autonomous vehicles, but that already put me head and shoulders above many of my contemporaries.
Now, my firm has weekly meetings called "Tesla Autopilot Review Team" and gets dozens of calls a month from across the country on incidents involving Autopilot or Full Self-Driving modes.
The vast majority of them involve minor or modest injuries that I'm simply incapable of reviewing. The barrier to entry due to the cost of litigation is so high.
In the Miami case, I spent over a million dollars on litigation costs alone — not including fees, but experts, depositions, data work, and everything else. I'm a contingency lawyer. The math on that has to be a $10-plus million case to make sense.
So we have to limit the cases that we look at to those involving catastrophic injuries or, unfortunately, a loss of life.
For someone who is minimally hurt: great for them, bad for their lawsuit. At the end of the day, the courthouse doors are closed to most of those people.
The Benavides case
The Benavides case was always about shared responsibility. We never tried to frame it as anything but that.
In every one of these Tesla Autopilot cases, there is typically an at-fault driver. Someone did something wrong. Someone failed to be an alert and attentive driver. That, however, should not absolve Tesla.
Tesla's frame was to focus on the last three seconds: What was the driver doing? What didn't they do? What systems were on? What systems were off?
The Benavides v. Tesla case involved a driver with Enhanced Autopilot enabled in his 2019 Tesla Model S.
Brett Schreiber
My role was to make it about something larger. Accidents happen, but systemic failures are caused by choices. If I made this case about the last three seconds, I would lose.
We can't change the facts about the driver's conduct. We talked about it. But we also argued that Tesla made decisions years before the collision — decisions to overstate, overhype, and oversell this technology, resulting in people over-relying on it just as this driver did.
So suddenly, it's not just about the driver's actions. What this driver did became inevitable, and that inevitability exists because of the choices Tesla made.
There is no playbook for AV lawsuits
In the world of auto-product liability, such as a tire-failure case or a seatback-failure case, there's a bit of a playbook. Meaning, a lot of the same experts are involved. Testing has been done.
The Benavides trial in Miami was only the third case ever to go to trial against Tesla, and the first time the plaintiffs won. We're in the very nascent stages of this type of litigation. We're building the plane as we're flying it.
In the last half-century of auto product-defect litigation, you could typically point to a mechanical failure mechanism. Some bolt failed, some hose disconnected, or a seatback bolt wasn't properly torqued, and as a result, the seat folded like a beach chair.
Here, if there are defect exists, they exist in the software. There is typically no mechanical thumbprint, no witness mark that shows you how the system failed. All of this is algorithmic. All of this lives in a world of ones and zeros.
The challenge is to turn algorithmic decision-making into something that we can discover and explain to jurors so they can understand how and why these crashes occurred.
Sometimes it's a camera-fusion failure. Sometimes it's a failure of automatic emergency braking or forward-collision warning to work. Sometimes a system activates in a place it should not have.
I don't need Tesla's source code to understand when a camera-fusion failure occurs. If three cameras target an issue and all three read it differently, I can show a jury that none of them could agree and there was no decision-maker.
The updates leave a glaring problem
There are a lot of Tesla vehicles still on the road today that use cameras and radar for the sensors.
People purchased these vehicles over the last decade with the idea or promise that they would reach a point where they would be fully self-driving and fully autonomous.
Elon Musk only recently said that not all Teslas will operate fully autonomously.
[In April, Musk said older Teslas with the Hardware 3 computer could not achieve unsupervised FSD without a hardware upgrade.]
Tesla has tried to push out over-the-air updates. We've seen in a number of areas that that's kind of like trying to push out an update for an iPhone 17 and make it fit on an iPhone 4. There are computing limitations.
Flash forward: We're now starting to see problems with the camera-onlysystem.
[There is a current National Highway Traffic Safety Administration investigation into Tesla's "degradation detection system." A filing from NHTSA's Office of Defects Investigation shows that the department is examining whether the system properly detects when conditions such as glare degrade camera visibility and warns drivers in time to respond.]
I believe in autonomy
I still fundamentally believe that autonomous driving is a tremendous societal good.
I believe that it is where the future of driving technology is headed, and it can be done well. This is not an indictment of the autonomous vehicle industry.
I think Tesla has to do two things.
One, the company has to be more intentional about the safety culture and about fixing bugs and problems far sooner.
The problem with an over-the-air software update is that it's one thing if I haven't updated my Spotify app. No one's going to lose their life,whether I'm running the most up-to-date version of Spotify.But someone can lose their life if they're not running the most up-to-date version of an autonomous vehicle software.
Number two, Tesla has to change the narrative that this is some glorious, full self-driving vehicle when it's not.
There is only one company in the world that markets partial automation as "full self-driving." The words partial and full in any reasonable interpretation of the English language do not mean the same thing.
I'm sorry, but calling it "Full Self-Driving Supervised" is not enough.
Two associates of Sam Bankman-Fried invested in Anthropic. After they were sentenced for fraud, their shares entered a legal black hole.
ANGELA WEIS/Getty Images; Tyler Le/BI
Sam Bankman-Fried's Anthropic stake was liquidated in bankruptcy court two years ago.
The feds seized shares belonging to two co-conspirators, Caroline Ellison and Nishad Singh.
The government quietly sold those Anthropic shares last year. It's unclear where the money will go.
A little over a year ago, the federal government acquired a stake in Anthropic without paying a penny.
The shares were seized from associates of the crypto fraudster Sam Bankman-Fried. From there, they disappeared into a legal black hole.
As Anthropic rockets toward what could be the biggest IPO in history, those shares could now be worth billions of dollars.
Bankman-Fried's own Anthropic stake was liquidated in the bankruptcy of his failed cryptocurrency exchange. The Anthropic equity owned by Caroline Ellison and Nishad Singh, two associates who invested alongside him, took a different path. The government seized those shares and sold them to existing Anthropic shareholders, according to a person familiar with the sale.
It's still unclear whether the proceeds will be paid out to victims of the $11 billion fraud case, or if the feds will just keep the money.
It also remains a mystery which investors bought Singh's and Ellison's stock and how much they paid, but Anthropic shares have been on such a dizzying climb that it's likely those investors made a killing on them.
Bankman-Fried was convicted in 2023 on fraud and money laundering charges and sentenced to 25 years in prison after the collapse of his cryptocurrency empire. Prosecutors said he used his crypto hedge fund, Alameda Research, to siphon billions of dollars from customers of FTX, his crypto exchange.
The story of the crypto criminals' Anthropic shares is a tale with extraordinary elements, including one of the biggest financial frauds in modern history, a messy bankruptcy docket, and a company that experienced one of the fastest appreciations in business history as the world suddenly realized AI's existential importance.
An intelligent investment
Bankman-Fried, Singh, and Ellison each invested in Anthropic's 2022 Series B funding round. Bankman-Fried bought $500 million worth, which according to court records represented 13.56% of Anthropic at the time. Singh acquired $40 million in shares, and Ellison acquired $10 million in shares, court records reviewed by Business Insider show.
The Anthropic shares purchased by Singh and Ellison could together be worth between $4.17 billion and $5.03 billion today, based on the $965 billion valuation the company announced this May, according to Olav Sorenson, who teaches venture capital strategy at UCLA. Harrison Rolfes, an analyst at PitchBook, put the combined figure at $2.62 billion. If Anthropic went public at a $2 trillion valuation, the shares would be worth about $5.44 billion, Rolfes said.
Among Sam Bankman-Fried's many investments was a sizable stake in Anthropic, the AI company that has since soared in value.
Bloomberg/Getty Images
Four of Bankman-Fried's close friends and executives at his companies pleaded guilty as co-conspirators. Two of them, Singh and Ellison, testified against him.
After the collapse of FTX, Bankman-Fried's Anthropic shares — which were owned by Clifton Bay, an entity affiliated with Alameda Research — ended up in bankruptcy court along with the other dredged-up remains of his companies. His Anthropic shares were liquidated to pay FTX's creditors.
The estate in 2024 sold Bankman-Fried's Anthropic shares to a few dozen buyers for a total of $1.3 billion, more than double what he paid. The largest stake went to an entity affiliated with the United Arab Emirates sovereign wealth fund, bankruptcy court filings show.
The feds take a stake in Anthropic
As part of their sentences, a judge required Singh and Ellison to forfeit their Anthropic shares, which prosecutors said could be considered proceeds of their crimes.
At Singh's sentencing hearing, one of his attorneys, Andrew Goldstein, said Singh purchased his shares before participating in the criminal conspiracy and "he actually may have had a legitimate legal claim" to the shares but agreed to give them up as part of his plea agreement "because it was the right thing to do."
Reached for comment for this story, Goldstein told Business Insider that Singh hopes the government is able to quickly distribute proceeds of the sale to FTX victims. An attorney for Ellison declined to comment.
Nishad Singh was required to give up his shares of Anthropic as part of his sentence.
Bloomberg/Getty Images
A federal judge ordered Ellison's and Singh's Anthropic shares to be transferred to the federal government, which took ownership of Ellison's shares in February 2025 and Singh's in April of that year, according to previously unreported court records.
Ordinarily, victims of crimes are compensated through a restitution process, which is overseen by courts. But the number of potential FTX victims could be in the millions, prosecutors said in court filings. As a result, the judge ruled, victim compensation would be handled through a process called remission, which is overseen by the Justice Department.
During Ellison's sentencing hearing, Justice Department prosecutors told the judge that the DOJ would either set up its own claims administration process to compensate victims, or work with the FTX bankruptcy process to identify victims and provide forfeited funds to them.
At the time, FTX's bankruptcy process was in full swing. A Delaware court appointed Sullivan & Cromwell, the elite Wall Street law firm, to untangle the company's assets, figure out who was owed money, and pay them.
The creditors in the FTX bankruptcy, prosecutors noted, largely overlapped with FTX's victims. They were generally depositors, lenders, and investors who were defrauded by Bankman-Fried and other executives. Prosecutors said the Justice Department could work with the bankruptcy estate's lawyers to get money back to them, as it had in previous large-scale financial frauds such as Bernie Madoff's Ponzi scheme.
There was an unusual twist that set the FTX bankruptcy apart. Bankman-Fried's investments — especially the Anthropic shares, along with some cryptocurrency — had grown substantially in value since FTX's collapse. Earlier in 2024, the Sullivan & Cromwell lawyers who'd taken over FTX projected that all the creditors would be repaid in full, with interest.
The Marshals take control
While prosecutors said they intended to use Singh's and Ellison's Anthropic shares for remission, the Justice Department could technically do whatever it wanted with them, according to Duncan Levin, a white-collar defense attorney who teaches a course on forfeiture at Harvard Law School.
"It's a very opaque process," he said. "It's completely at the discretion, by law, of the attorney general of the United States."
Typically, when the feds seize shares of private companies through criminal asset forfeiture, they send the shares to the US Marshals Service Complex Assets Unit for liquidation. The unit tries to value the shares as any other investor would, said Michael Bachner, a white-collar criminal and securities litigation attorney.
"They may look to: What would an institutional purchaser pay for these securities?" Bachner said. "Are there funds that are valuing the securities? Is there a secondary market already out there?"
For the sale of the Anthropic shares, timing was crucial. The Marshals Service is supposed to preserve as much value as possible, Bachner said. By the time the feds got hold of the shares, Anthropic had become an economy-shifting AI giant. In its March 2025 Series E fundraising round, Anthropic was worth $61.5 billion. By its Series G round at the start of 2026, it was worth $380 billion.
At the same time, each fundraising round diluted the Series B shares. The FTX estate told the bankruptcy court that Bankman-Fried's shares, which in 2022 represented 13.56% of Anthropic, represented 7.84% of the AI company in January 2024.
Sunil Kavuri, a victim of Sam Bankman-Fried who has advocated for FTX creditors, said it would be "diabolical" for the government to hold onto the proceeds.
David Dee Delgado/Getty Images
The Marshals Service sold Singh's and Ellison's shares to existing Anthropic investors sometime last year, according to the person with knowledge of the sale.
It's not clear when exactly the agency sold the shares, to which investors, how those investors were chosen, at what price the shares were sold, or how much money the government made in the sale. Depending on when they were sold in 2025, the combined shares could have been worth between $300 million and $1.1 billion, according to Sorenson, the UCLA professor. Rolfes, the PitchBook analyst, estimated a range between $250 million and $630 million, depending on the timing.
The Marshals Service declined to comment. A representative for the Justice Department said information about asset sales and victim compensation is confidential.
The revenue from the sale of Singh's and Ellison's Anthropic shares doesn't appear to have been transferred to the FTX estate as of the end of June this year, according to bankruptcy court filings from the estate, which continues to compensate victims and creditors.
The FTX estate received $638 million last year from assets seized by the Justice Department, according to the estate's annual report for 2025. Other filings show that nearly all of that amount came from the sale of Robinhood shares previously owned by Bankman-Fried. The estate expects to receive about $400 million more from the government at some point in the future, according to the annual report. That would include proceeds from cryptocurrency and other investments made by Bankman-Fried.
One of Bankman-Fried's victims, Sunil Kavuri, told Business Insider that the Justice Department should use the proceeds of Ellison's and Singh's Anthropic shares to compensate victims. It would be "diabolical" for the government to hold onto the proceeds, he said.
Victims haven't actually been made whole, he said, because the bankruptcy court calculated the losses of FTX depositors using the time of FTX's bankruptcy declaration, when crypto prices were at a low ebb. Kavuri previously argued in bankruptcy court that the FTX estate should have held on to its Anthropic shares for longer to take advantage of the company's swift growth.
Representatives for the FTX estate declined to comment. The Justice Department spokesperson said the matter was ongoing, and that the DOJ "prioritizes victim compensation from forfeiture and takes all steps to ensure forfeited funds are provided to victims."
The government could end up just keeping the money, Bachner said.
"They've invested millions and millions and millions of dollars in prosecuting Bankman-Fried. And they want to get at least reimbursed for their costs, so sometimes they'll do that," he said. "It's really a unilateral government decision."
Jack Newsham and Katie Roof contributed reporting for this story.
