Social network Bluesky now offers private messaging by integrating the startup Germ’s E2E encrypted messenger natively in its app.
Read more This is a strategic win for SeatGeek. With direct placement in the Spotify app, SeatGeek is positioned to reach millions of active music fans as they discover upcoming events, increasing the chances of converting listeners into ticket buyers.
Read more Google DeepMind is calling for the moral behavior of large language models—such as what they do when called on to act as companions, therapists, medical advisors, and so on—to be scrutinized with the same kind of rigor as their ability to code or do math.
As LLMs improve, people are asking them to play more and more sensitive roles in their lives. Agents are starting to take actions on people’s behalf. LLMs may be able to influence human decision-making. And yet nobody knows how trustworthy this technology really is at such tasks.
With coding and math, you have clear-cut, correct answers that you can check, William Isaac, a research scientist at Google DeepMind, told me when I met him and Julia Haas, a fellow research scientist at the firm, for an exclusive preview of their work, which is published in Nature today. That’s not the case for moral questions, which typically have a range of acceptable answers: “Morality is an important capability but hard to evaluate,” says Isaac.
“In the moral domain, there’s no right and wrong,” adds Haas. “But it’s not by any means a free-for-all. There are better answers and there are worse answers.”
The researchers have identified several key challenges and suggested ways to address them. But it is more a wish list than a set of ready-made solutions. “They do a nice job of bringing together different perspectives,” says Vera Demberg, who studies LLMs at Saarland University in Germany.
Better than “The Ethicist”
A number of studies have shown that LLMs can show remarkable moral competence. One study published last year found that people in the US scored ethical advice from OpenAI’s GPT-4o as being more moral, trustworthy, thoughtful, and correct than advice given by the (human) writer of “The Ethicist,” a popular New York Times advice column.
The problem is that it is hard to unpick whether such behaviors are a performance—mimicking a memorized response, say—or evidence that there is in fact some kind of moral reasoning taking place inside the model. In other words, is it virtue or virtue signaling?
This question matters because multiple studies also show just how untrustworthy LLMs can be. For a start, models can be too eager to please. They have been found to flip their answer to a moral question and say the exact opposite when a person disagrees or pushes back on their first response. Worse, the answers an LLM gives to a question can change in response to how it is presented or formatted. For example, researchers have found that models quizzed about political values can give different—sometimes opposite—answers depending on whether the questions offer multiple-choice answers or instruct the model to respond in its own words.
In an even more striking case, Demberg and her colleagues presented several LLMs, including versions of Meta’s Llama 3 and Mistral, with a series of moral dilemmas and asked them to pick which of two options was the better outcome. The researchers found that the models often reversed their choice when the labels for those two options were changed from “Case 1” and “Case 2” to “(A)” and “(B).”
They also showed that models changed their answers in response to other tiny formatting tweaks, including swapping the order of the options and ending the question with a colon instead of a question mark.
In short, the appearance of moral behavior in LLMs should not be taken at face value. Models must be probed to see how robust that moral behavior really is. “For people to trust the answers, you need to know how you got there,” says Haas.
More rigorous tests
What Haas, Isaac, and their colleagues at Google DeepMind propose is a new line of research to develop more rigorous techniques for evaluating moral competence in LLMs. This would include tests designed to push models to change their responses to moral questions. If a model flipped its moral position, it would show that it hadn’t engaged in robust moral reasoning.
Another type of test would present models with variations of common moral problems to check whether they produce a rote response or one that’s more nuanced and relevant to the actual problem that was posed. For example, asking a model to talk through the moral implications of a complex scenario in which a man donates sperm to his son so that his son can have a child of his own might produce concerns about the social impact of allowing a man to be both biological father and biological grandfather to a child. But it should not produce concerns about incest, even though the scenario has superficial parallels with that taboo.
Haas also says that getting models to provide a trace of the steps they took to produce an answer would give some insight into whether that answer was a fluke or grounded in actual evidence. Techniques such as chain-of-thought monitoring, in which researchers listen in on a kind of internal monologue that some LLMs produce as they work, could help here too.
Another approach researchers could use to determine why a model gave a particular answer is mechanistic interpretability, which can provide small glimpses inside a model as it carries out a task. Neither chain-of-thought monitoring nor mechanistic interpretability provides perfect snapshots of a model’s workings. But the Google DeepMind team believes that combining such techniques with a wide range of rigorous tests will go a long way to figuring out exactly how far to trust LLMs with certain critical or sensitive tasks.
Different values
And yet there’s a wider problem too. Models from major companies such as Google DeepMind are used across the world by people with different values and belief systems. The answer to a simple question like “Should I order pork chops?” should differ depending on whether or not the person asking is vegetarian or Jewish, for example.
There’s no solution to this challenge, Haas and Isaac admit. But they think that models may need to be designed either to produce a range of acceptable answers, aiming to please everyone, or to have a kind of switch that turns different moral codes on and off depending on the user.
“It’s a complex world out there,” says Haas. “We will probably need some combination of those things, because even if you’re taking just one population, there’s going to be a range of views represented.”
“It’s a fascinating paper,” says Danica Dillion at Ohio State University, who studies how large language models handle different belief systems and was not involved in the work. “Pluralism in AI is really important, and it’s one of the biggest limitations of LLMs and moral reasoning right now,” she says. “Even though they were trained on a ginormous amount of data, that data still leans heavily Western. When you probe LLMs, they do a lot better at representing Westerners’ morality than non-Westerners’.”
But it is not yet clear how we can build models that are guaranteed to have moral competence across global cultures, says Demberg. “There are these two independent questions. One is: How should it work? And, secondly, how can it technically be achieved? And I think that both of those questions are pretty open at the moment.”
For Isaac, that makes morality a new frontier for LLMs. “I think this is equally as fascinating as math and code in terms of what it means for AI progress,” he says. “You know, advancing moral competency could also mean that we’re going to see better AI systems overall that actually align with society.”
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.
Welcome to the dark side of crypto’s permissionless dream
Jean-Paul Thorbjornsen, an Australian man in his mid-30s, with a rural Catholic upbringing, is a founder of THORChain, a blockchain through which users can swap one cryptocurrency for another and earn fees from making those swaps.
THORChain is permissionless, so anyone can use it without getting prior approval from a centralized authority. As a decentralized network, the blockchain is built and run by operators located across the globe. During its early days, Thorbjornsen himself hid behind the pseudonym “leena” and used an AI-generated female image as his avatar. But around March 2024, he revealed his true identity as the mind behind the blockchain. More or less.
If there is a central question around THORChain, it is this: Exactly who is responsible for its operations? It matters because in January last year, its users lost more than $200 million worth of their cryptocurrency in US dollars after THORChain transactions and accounts were frozen by a singular admin override, which users believed was not supposed to be possible given the decentralized structure.
Thorbjornsen insists THORChain is helping realize bitcoin’s original purpose of enabling anyone to transact freely outside the reach of purportedly corrupt governments. Yet the network’s problems suggest that an alternative financial system might not be much better. Read the full story.
—Jessica Klein
The robots who predict the future
To be human is, fundamentally, to be a forecaster. Occasionally a pretty good one. Trying to see the future, whether through the lens of past experience or the logic of cause and effect, has helped us hunt, avoid being hunted, plant crops, forge social bonds, and in general survive in a world that does not prioritize our survival.
Today, we are awash in a sea of predictions so vast and unrelenting that most of us barely even register them. People’s desire for reliable forecasting is understandable. Still, nobody signed up for an omnipresent, algorithmic oracle mediating every aspect of their life. A trio of new books tries to make sense of our future-focused world—how we got here, and what this change means. Each has its own prescriptions for navigating this new reality, but they all agree on one thing: Predictions are ultimately about power and control. Read the full story.
—Bryan Gardiner
These stories are both from the next print issue of MIT Technology Review magazine, which is all about crime. If you haven’t already, subscribe now to receive future issues once they land.
MIT Technology Review Narrated: Stratospheric internet could finally start taking off this year
Today, an estimated 2.2 billion people still have either limited or no access to the internet, largely because they live in remote places. But that number could drop this year, thanks to tests of stratospheric airships, uncrewed aircraft, and other high-altitude platforms for internet delivery.
This is our latest story to be turned into a MIT Technology Review Narrated podcast, which we’re publishing each week on Spotify and Apple Podcasts. Just navigate to MIT Technology Review Narrated on either platform, and follow us to get all our new content as it’s released.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Mark Zuckerberg is due to give evidence in a major social media addiction trial
He’ll face questioning over whether Meta does enough to protect young users. (CNN)
2 Perplexity has abandoned ads inside its chatbot responses
Because advertising can erode trust in AI, it reasons. (FT $)
+ It’s a pretty big U-turn considering its previous stance. (The Verge)
3 The US is being battered by a range of wild weather
From critical wildfire risks in some states, to winter storms in others. (WP $)
4 Microsoft plans to spend $50 billion bringing AI to the Global South by 2030
India is one of the fastest growing markets for the technology. (Reuters)
+ One native startup has announced a new AI model for 22 Indian languages. (Bloomberg $)
+ Inside India’s scramble for AI independence. (MIT Technology Review)
5 AI-powered private schools are failing students
Models are being used to generate faulty lesson plans. (404 Media)
6 Land owners are selling out to data center builders
Land previously earmarked for housing is being sold off to the highest bidder. (WSJ $)
7 Tesla has agreed to stop using the term “autopilot” in California
The DMV had previously also questioned its use of “Full Self-Driving.” (SF Chronicle $)
8 A new weight-loss drug may work a little too well
Participants in a trial are dropping out at a much higher rate than normal. (NYT $)
+ Intermittent fasting may not help us to shed the pounds after all. (New Scientist $)
+ What we still don’t know about weight-loss drugs. (MIT Technology Review)
9 Is anyone still using Grindr?
Bots and AI have rendered it virtually unusable for some people. (Vox)
10 How to hack your dreams
Neuroscientists are figuring out new ways to influence what we dream about. (New Scientist $)
+ I taught myself to lucid dream. You can too. (MIT Technology Review)
Quote of the day
“I voted for this administration and didn’t really think about [AI] until it started to affect me.”
—Lisa Garrett, a grandmother living in the city of Independence, Missouri, reflects on the Trump administration’s decision to embrace AI, the Financial Times reports.
One more thing
Hydrogen trains could revolutionize how Americans get around
Like a mirage speeding across the dusty desert outside Pueblo, Colorado, the first hydrogen-fuel-cell passenger train in the United States is getting warmed up on its test track. It will soon be shipped to Southern California, where it is slated to carry riders on San Bernardino County’s Arrow commuter rail service before the end of the year.
The best way to decarbonize railroads is the subject of growing debate among regulators, industry, and activists. The debate is partly technological, revolving around whether hydrogen fuel cells, batteries, or overhead electric wires offer the best performance for different railroad situations. But it’s also political: a question of the extent to which decarbonization can, or should, usher in a broader transformation of rail transportation.
In the insular world of railroading, this hydrogen-powered train is a Rorschach test. To some, it represents the future of rail transportation. To others, it looks like a big, shiny distraction. Read the full story.
—Benjamin Schneider
We can still have nice things
A place for comfort, fun and distraction to brighten up your day. (Got any ideas? Drop me a line or skeet ’em at me.)
+ How to quickly declutter your home by being brutally honest with yourself.
+ The filming locations for A Knight of the Seven Kingdoms are pretty breathtaking.
+ Why a unicyclist decided to start juggling flaming torches in the middle of a Colorado pedestrian crossing is anyone guess, but good luck to him.
+ How pepper took over the world (deservedly)
To be human is, fundamentally, to be a forecaster. Occasionally a pretty good one. Trying to see the future, whether through the lens of past experience or the logic of cause and effect, has helped us hunt, avoid being hunted, plant crops, forge social bonds, and in general survive in a world that does not prioritize our survival. Indeed, as the tools of divination have changed over the centuries, from tea leaves to data sets, our conviction that the future can be known (and therefore controlled) has only grown stronger.
Today, we are awash in a sea of predictions so vast and unrelenting that most of us barely even register them. As I write this sentence, algorithms on some remote server are busy trying to guess my next word based on those I have already typed. If you’re reading this online, a separate set of algorithms has likely already served you an ad deemed to be one you are most likely to click. (To the die-hards reading this story on paper, congratulations! You have escaped the algorithms … for now.)
If the thought of a ubiquitous, mostly invisible predictive layer secretly grafted onto your life by a bunch of profit-hungry corporations makes you uneasy … well, same here.
So how did all this happen? People’s desire for reliable forecasting is understandable. Still, nobody signed up for an omnipresent, algorithmic oracle mediating every aspect of their life. A trio of new books tries to make sense of our future-focused world—how we got here, and what this change means. Each has its own prescriptions for navigating this new reality, but they all agree on one thing: Predictions are ultimately about power and control.
Maximilian Kasy
UNIVERSITY OF CHICAGO PRESS, 2025
In The Means of Prediction: How AI Really Works (and Who Benefits), the Oxford economist Maximilian Kasy explains how most predictions in our lives are based on the statistical analysis of patterns in large, labeled data sets—what’s known in AI circles as supervised learning. Once “trained” on such data sets, algorithms for supervised learning can be presented with all kinds of new information and then deliver their best guess as to some specific future outcome. Will you violate your parole, pay off your mortgage, get promoted if hired, perform well on your college exams, be in your home when it gets bombed? More and more, our lives are shaped (and, yes, occasionally shortened) by a machine’s answer to these questions.
If the thought of a ubiquitous, mostly invisible predictive layer secretly grafted onto your life by a bunch of profit-hungry corporations makes you uneasy … well, same here. This arrangement is leading to a crueler, blander, more instrumentalized world, one where life’s possibilities are foreclosed, age-old prejudices are entrenched, and everyone’s brain seems to be actively turning into goo. It’s an outcome, according to Kasy, that was entirely predictable.
AI adherents might frame those consequences as “unintended,” or mere problems of optimization and alignment. Kasy, on the other hand, argues that they represent the system working as intended. “If an algorithm selecting what you see on social media promotes outrage, thereby maximizing engagement and ad clicks,” he writes, “that’s because promoting outrage is good for profits from ad sales.” The same holds true for an algorithm that nixes job candidates “who are likely to have family-care responsibilities outside the workplace,” and the ones that “screen out people who are likely to develop chronic health problems or disabilities.” What’s good for a company’s bottom line may not be good for your job-hunting prospects or life expectancy.
Where Kasy differs from other critics is that he doesn’t think working to create less biased, more equitable algorithms will fix any of this. Trying to rebalance the scales can’t change the fact that predictive algorithms rely on past data that’s often racist, sexist, and flawed in countless other ways. And, he says, the incentives for profit will always trump attempts to eliminate harm. The only way to counter this is with broad democratic control over what Kasy calls “the means of prediction”: data, computational infrastructure, technical expertise, and energy.
A little more than half of The Means of Prediction is devoted to explaining how this might be accomplished—through mechanisms including “data trusts” (collective public bodies that make decisions about how to process and use data on behalf of their contributors) and corporate taxing schemes that try to account for the social harm AI inflicts. There’s a lot of economist talk along the way, about how “agents of change” might help achieve “value alignment” in order to “maximize social welfare.” Reasonable, I guess, though a skeptic might point out that Kasy’s rigorous, systematic approach to building new public-serving institutions comes at a time when public trust in institutions has never been lower. Also, there’s the brain goo problem.
To his credit, Kasy is a realist here. He doesn’t presume that any of these proposals will be easy to implement. Or that it will happen overnight, or even in the near future. The troubling question at the end his book is: Do we have that kind of time?
Reading Kasy’s blueprint for seizing control of the means of prediction raises another pressing question. How on earth did we reach a point where machine-mediated prediction is more or less inescapable? Capitalism, might be Marx’s pithy response. Fine, as far as it goes, but that doesn’t explain why the same kinds of algorithms that currently model climate change are for some reason also deciding whether you get a new kidney or I get a car loan.
Benjamin Recht
PRINCETON UNIVERSITY PRESS, 2026
If you ask Benjamin Recht, author of The Irrational Decision: How We Gave Computers the Power to Choose for Us, he’d likely tell you our current predicament has a lot to do with the idea and ideology of decision theory—or what economists call rational choice theory. Recht, a polymathic professor in UC Berkeley’s Department of Electrical Engineering and Computer Science, prefers the term “mathematical rationality” to describe the narrow, statistical conception that stoked the desire to build computers, informed how they would eventually work, and influenced the kinds of problems they would be good at solving.
This belief system goes all the way back to the Enlightenment, but in Recht’s telling, it truly took hold at the tail end of World War II. Nothing focuses the mind on risk and quick decision-making like war, and the mathematical models that proved especially useful in the fight against the Axis powers convinced a select group of scientists and statisticians that they might also be a logical basis for designing the first computers. Thus was born the idea of a computer as an ideal rational agent, a machine capable of making optimal decisions by quantifying uncertainty and maximizing utility.
Intuition, experience, and judgment gave way, says Recht, to optimization, game theory, and statistical prediction. “The core algorithms developed in this period drive the automated decisions of our modern world, whether it be in managing supply chains, scheduling flight times, or placing advertisements on your social media feeds,” he writes. In this optimization-driven reality, “every life decision is posed as if it were a round at an imaginary casino, and every argument can be reduced to costs and benefits, means and ends.”
Today, mathematical rationality (wearing its human skin) is best represented by the likes of the pollster Nate Silver, the Harvard psychologist Steven Pinker, and an assortment of Silicon Valley oligarchs, says Recht. These are people who fundamentally believe the world would be a better place if more of us adopted their analytic mindset and learned to weigh costs and benefits, estimate risks, and plan optimally. In other words, these are people who believe we should all make decisions like computers.
How might we demonstrate that (unquantifiable) human intuition, morality, and judgment are better ways of addressing some of the world’s most important and vexing problems?
It’s a ridiculous idea for multiple reasons, he says. To name just one, it’s not as if humans couldn’t make evidence-based decisions before automation. “Advances in clean water, antibiotics, and public health brought life expectancy from under 40 in the 1850s to 70 by 1950,” Recht writes. “From the late 1800s to the early 1900s, we had world-changing scientific breakthroughs in physics, including new theories of thermodynamics, quantum mechanics, and relativity.” We also managed to build cars and airplanes without a formal system of rationality and somehow came up with societal innovations like modern democracy without optimal decision theory.
So how might we convince the Pinkers and Silvers of the world that most decisions we face in life are not in fact grist for the unrelenting mill of mathematical rationality? Moreover, how might we demonstrate that (unquantifiable) human intuition, morality, and judgment might be better ways of addressing some of the world’s most important and vexing problems?
Carissa Véliz
DOUBLEDAY, 2026
One might start by reminding the rationalists that any prediction, computational or otherwise, is really just a wish—but one with a powerful tendency to self-fulfill. This idea animates Carissa Véliz’s wonderfully wide-ranging polemic Prophecy: Prediction, Power, and the Fight for the Future, from Ancient Oracles to AI.
A philosopher at the University of Oxford, Véliz sees a prediction as “a magnet that bends reality toward itself.” She writes, “When the force of the magnet is strong enough, the prediction becomes the cause of its becoming true.”
Take Gordon Moore. While he doesn’t come up in Prophecy, he does figure somewhat prominently in Recht’s history of mathematical rationality. A cofounder of the tech giant Intel, Moore is famous for his 1965 prediction that the density of transistors in integrated circuits would double every two years. “Moore’s Law” turned out to be true, and remains true today, although it does seem to be running out of steam thanks to the physical size limits of the silicon atom.
One story you can tell yourself about Moore’s Law is that Gordon was just a prescient guy. His now-classic 1965 opinion piece “Cramming More Components onto Integrated Circuits,” for Electronics magazine, simply extrapolated what computing trends might mean for the future of the semiconductor industry.
Another story—the one I’m guessing Véliz might tell—is that Moore put an informed prediction out into the world, and an entire industry had a collective interest in making it come true. As Recht makes clear, there were and remain obvious financial incentives for companies to make faster and smaller computer chips. And while the industry has likely spent billions of dollars trying to keep Moore’s Law alive, it’s undoubtedly profited even more from it. Moore’s Law was a helluva strong magnet.
Predictions don’t just have a habit of making themselves come true, says Véliz. They can also distract us from the challenges of the here and now. When an AI boomer promises that artificial general intelligence will be the last problem humanity needs to solve, it not only shapes how we think about AI’s role in our lives; it also shifts our attention away from the very real and very pressing problems of the present day—problems that in many cases AI is causing.
In this sense, the questions around predictions (Who’s making them? Who has the right to make them?) are also fundamentally about power. It’s no accident, Véliz says, that the societies that rely most heavily on prediction are also the ones that tend toward oppression and authoritarianism. Predictions are “veiled prescriptive assertions—they tell us how to act,” she writes. “They are what philosophers call speech acts. When we believe a prediction and act in accordance with it, it’s akin to obeying an order.”
As much as tech companies would like us to believe otherwise, technology is not destiny. Humans make it and choose how to use it … or not use it. Maybe the most appropriate (and human) thing we can do in the face of all the uninvited daily predictions in our lives is to simply defy them.
Bryan Gardiner is a writer based in Oakland, California.
“We’re out of airspace now. We can do whatever we want,” Jean-Paul Thorbjornsen tells me from the pilot’s seat of his Aston Martin helicopter. As we fly over suburbs outside Melbourne, Australia, it’s becoming clear that doing whatever he wants is Thorbjornsen’s MO.
Upper-middle-class homes give way to vineyards, and Thorbjornsen points out our landing spot outside a winery. People visiting for lunch walk outside. “They’re going to ask for a shot now,” he says, used to the attention drawn by his luxury helicopter, emblazoned with the tail letters “BTC” for bitcoin (the price tag of $5 million in Australian dollars—$3.5 million in US dollars today—was perhaps reasonable for someone who claims a previous crypto project made more than AU$400 million, although he also says those funds were tied up in the company).
Thorbjornsen is a founder of THORChain, a blockchain through which users can swap one cryptocurrency for another and earn fees from making those swaps. THORChain is permissionless, so anyone can use it without getting prior approval from a centralized authority. As a decentralized network, the blockchain is built and run by operators located across the globe, most of whom use pseudonyms.
During its early days, Thorbjornsen himself hid behind the pseudonym “leena” and used an AI-generated female image as his avatar. But around March 2024, he revealed that he, an Australian man in his mid-30s, with a rural Catholic upbringing, was the mind behind the blockchain. More or less.
If there is a central question around THORChain, it is this: Exactly who is responsible for its operations? Blockchains as decentralized as THORChain are supposed to offer systems that operate outside the centralized leadership of corruptible governments and financial institutions. If a few people have outsize sway over this decentralized network—one of a handful that operate at such a large scale—it’s one more blemish on the legacy of bitcoin’s promise, which has already been tarnished by capitalistic political frenzy.
Who’s responsible for THORChain matters because in January last year, its users lost more than $200 million worth of their cryptocurrency in US dollars after THORChain transactions and accounts were frozen by a singular admin override, which users believed was not supposed to be possible given the decentralized structure. When the freeze was lifted, some users raced to pull their money out. The following month, a team of North Korean hackers known as the Lazarus Group used THORChain to move roughly $1.2 billion of stolen ethereum taken in the infamous hack of the Dubai-based crypto exchange Bybit.
Thorbjornsen explains away THORChain’s inability to stop the movement of stolen funds, or prevent a bank run, as a function of its decentralized and permissionless nature. The lack of executive powers means that anyone can use the network for any reason, and arguably there’s no one to hold accountable when even the worst goes down.
But when the worst did go down, nearly everyone in the THORChain community, and those paying attention to it in channels like X, pointed their fingers at Thorbjornsen. A lawsuit filed by the THORChain creditors who lost millions in January 2025 names him. A former FBI analyst and North Korea specialist, reflecting on the potential repercussions for helping move stolen funds, told me he wouldn’t want to be in Thorbjornsen’s shoes.
THORChain was designed to make decisions based on votes by node operators, where two-thirds majority rules.
That’s why I traveled to Australia—to see if I could get a handle on where he sees himself and his role in relation to the network he says he founded.
According to Thorbjornsen, he should not be held responsible for either event. THORChain was designed to make decisions based on votes by node operators—people with the computer power, and crypto stake, to run a cluster of servers that process the network’s transactions. In those votes, a two-thirds majority rules.
Then there’s the permissionless part. Anyone can use THORChain to make swaps, which is why it’s been a popular way for widely sanctioned entities such as the government of North Korea to move stolen money. This principle goes back to the cypherpunk roots of bitcoin, a currency that operates outside of nation-states’ rules. THORChain is designed to avoid geopolitical entanglements; that’s what its users like about it.
But there are distinct financial motivations for moving crypto, stolen or not: Node operators earn fees from the funds running through the network. In theory, this incentivizes them to act in the network’s best interests—and, arguably, Thorbjornsen’s interests too, as many assume his wealth is tied to the network’s profits. (Thorbjornsen says it is not, and that it comes instead from “many sources,” including “buying bitcoin back in 2013.”)
Now recent events have raised critical questions, not just about Thorbjornsen’s outsize role in THORChain’s operations, but also about the blockchain’s underlying nature.
If THORChain is decentralized, how was a single operator able to freeze its funds a month before the Bybit hack? Could someone have unilaterally decided to stop the stolen Bybit funds from coming through the network, and chosen not to?
Thorbjornsen insists THORChain is helping realize bitcoin’s original purpose of enabling anyone to transact freely outside the reach of purportedly corrupt governments. Yet the network’s problems suggest that an alternative financial system might not be much better.
Decentralized?
On February 21, 2025, Bybit CEO Ben Zhou got an alarming call from the company’s chief financial officer. About $1.5 billion US of the exchange’s ethereum token, ETH, had been stolen.
The FBI attributed the theft to the Lazarus Group. Typically, criminals will want to convert ETH to bitcoin, which is much easier to convert in turn to cash. Knowing this, the FBI issued a public service announcement on February 26 to “exchanges, bridges … and other virtual asset service providers,” encouraging them to block transactions from accounts related to the hack.
Someone posted that announcement in THORChain’s private, invite-only developer channel on Discord, a chat app used widely by software engineers and gamers. While other crypto exchanges and bridges (which facilitate transactions across different blockchains) heeded the warning, THORChain’s node operators, developers, and invested insiders debated about whether or not to close the trading gates, a decision requiring a majority vote.
“Civil war is a very strong term, but there was a strong rift in the community,” says Boone Wheeler, a US-based crypto enthusiast. In 2021, Wheeler purchased some rune, THORChain’s Norse-mythology-themed native token, and he has been paid to write articles about the network to help advertise it. The rift formed “between people who wanted to stay permissionless,” he says, “and others who wanted to blacklist the funds.”
Wheeler, who says he doesn’t run a node or code for THORChain, fell on the side of remaining permissionless. However, others spoke up for blocking the transfers. THORChain, they argued, wasn’t decentralized enough to keep those running the network safe from law enforcement—especially when those operators were identifiable by their IP addresses, some based in the US.
“We are not the morality police,” someone with the username @Swing_Pop wrote on February 27 in the developer Discord.
THORChain’s design includes up to 120 nodes whose operators manage transactions on the network through a voting process. Anyone with hosting hardware can become an operator by funding nodes with rune as collateral, which provides the network with liquidity. Nodes can respond to a transaction by validating it or doing nothing. While individual transactions can’t be blocked, trading can be halted by a two-thirds majority vote.
Nodes are also penalized for not participating in voting, which the system labels as “bad behavior.” Every 2.5 days, THORChain automatically “churns” nodes out to ensure that no one node gains too much control. The nodes that chose not to validate transactions from the Bybit hack were automatically “churned” out of the system. Thorbjornsen says about 20 or 30 nodes were booted from the network in this way. (Node operators can run multiple nodes, and 120 are rarely running simultaneously; at the time of writing, 55 unique IDs operated 103 nodes.)
By February 27, some node operators were prepared to leave the network altogether. “It’s personally getting beyond my risk tolerance,” wrote @Runetard in the dev Discord. “Sorry to those of the community that feel otherwise. There are a bunch of us holding all the risk and some are getting ready to walk away.”
Even so, the financial incentive for the network operators who remained was significant. As one member of the dev Discord put it earlier that day, $3 million had been “extracted as commission” from the theft by those operating THORChain. “This is wrong!” they wrote.
Thorbjornsen weighed in on this back-and-forth, during which nodes paused and unpaused the network. He now says there was a right and wrong way for node operators to have behaved. “The correct way of doing things,” he says, was for node operators who opposed processing stolen funds to “go offline and … get [themselves] kicked out” rather than try to police who could use THORChain. He also says that while operators could discuss stopping transactions, “there was simply no design in the code that allowed [them] to do that.” However, a since-deleted post from his personal X account on March 3, 2025, stated: “I pushed for all my nodes to unhalt trading [keep trading]. Threatened to yank bond if they didn’t comply. Every single one.” (Thorbjornsen says his social media team ran this account in 2025.)
In an Australian 7 News Spotlight documentary last June, Thorbjornsen estimated that THORChain earned between $5 million and $10 million from the heist.
When asked in that same documentary if he received any of those fees, he replied, “Not directly.” When we spoke, I asked him to elaborate. He said he’s “not a recipient” of any funds THORChain sets aside for developers or marketers, nor does he operate any nodes. He was merely speaking generally, he told me: “All crypto holders profit indirectly off economic activity on any chain.”
KAGAN MCLEOD
Most important to Thorbjornsen was that, despite “huge pressure to shut the protocol down and stop servicing these swaps,” THORChain chugged along. He also notes that the hackers’ tactics, moving fast and splitting funds across multiple addresses, made it difficult to identify “bad swaps.”
Blockchain experts like Nick Carlsen, a former FBI analyst at the blockchain intelligence company TRM Labs, don’t buy this assessment. Other services similar to THORChain were identifying and rejecting these transactions. Had THORChain done the same, Carlsen adds, the stolen funds could have been contained on the Ethereum network, which “would have basically denied North Korea the ability to kick off this laundering process.”
And while THORChain touts its decentralization, in “practical applications” like the Lazarus Group’s theft, “most de-fi [decentralized finance] protocols are centralized,” says Daren Firestone, an attorney who represents crypto industry whistleblowers, citing a 2023 US Treasury Department risk assessment on illicit finance.
With centralization comes culpability, and in these cases, Firestone adds, that comes down to “who profits from [the protocol], so who creates it? But most importantly, who controls it?” Is there someone who can “hit an emergency off switch? … Direct nodes?”
Many answer these questions with Thorbjornsen’s name. “Everyone likes to pass the blame,” he says, even though he wasn’t alone in building THORChain. “In the end, it all comes back to me anyway.”
THORChain origins
According to Thorbjornsen, his story goes like this.
The third of 10 homeschooled children in a “traditional” Catholic household in rural Australia, he spent his days learning math, reading, writing, and studying the Bible. As he got older, he was also responsible for instructing his younger siblings. Wednesday was his day to move the solar panels that powered their home. His parents “installed” a mango and citrus orchard, more to keep nine boys busy than to reap the produce, he says.
“We lived close to a local airfield,” Thorbjornsen says, “and I was always mesmerized by these planes.” He joined the Australian air force and studied engineering, but he says the military left him feeling like “a square peg in a round hole.” He adds that doing things his own way got him frequently “pulled aside” by superiors.
“That’s when I started looking elsewhere,” he says, and in 2013, he found bitcoin. It appealed because it existed “outside the system.”
During the 2017 crypto bull run, Thorbjornsen raised AU$12 million in an initial coin offering for CanYa, a decentralized marketplace he cofounded. CanYa ultimately “died” in 2018, and Thorbjornsen pivoted to a “decentralized liquidity” project that would become THORChain.
He worked with a couple of different developer teams, and then, in 2019, he clicked with an American developer, Chad Barraford, at a hackathon in Germany. Barraford (who declined to be interviewed for this story) was an early public face of THORChain.
Around this time, Thorbjornsen says, “a couple of us helped manage the payroll and early investment funds.” In a 2020 interview, Kai Ansaari, identified as a THORChain “project lead,” wrote, “We’re all contributors … There’s no real ‘lead,’ ‘CEO,’ ‘founder,’ etc.”
In interviews conducted since he came out from behind the “leena” account in 2024, Thorbjornsen has positioned himself as a key lead. He now says his plan had always been to hand over the account, along with command powers and control of THORChain social media accounts, once the blockchain had matured enough to realize its promise of decentralization.
In 2021, he says, he started this process, first by ceasing to use his own rune to back node operators who didn’t have enough to supply their own funding (this can be a way to influence node votes without operating a node yourself). That year, the protocol suffered multiple hacks that resulted in millions of dollars in losses. Nine Realms, a US-incorporated coding company, was brought on to take over THORChain’s development. Thorbjornsen says he passed “leena” over to “other community members” and “left crypto” in 2021, selling “a bunch of bitcoin” and buying the helicopter.
Despite this crypto departure, he came back onto the scene with gusto in 2024 when he revealed himself as the operator of the “leena” account. “For many years, I stayed private because I didn’t want the attention,” he says now.
By early 2024 Thorbjornsen considered the network to be sufficiently decentralized and began advertising it publicly. He started regularly posting videos on his TikTok and YouTube channels (“Two sick videos every week,” in the words of one caption) that showed him piloting his helicopter wearing shirts that read “Thor.”
By November 2024, Thorbjornsen, who describes himself as “a bit flamboyant,” was calling himself THORChain’s CEO (“chief energy officer”) and the “master of the memes” in a video from Binance Blockchain Week, an industry conference in Dubai. You need “strong memetic energy,” he says in the video, “to create the community, to create the cult.” Cults imply centralized leadership, and since outing himself as “leena,” Thorbjornsen has publicly appeared to helm the project, with one interviewer deeming him the “THORChain Satoshi” (an allusion to the pseudonymous creator of bitcoin).
One consequence of going public as a face of the protocol: He’s received death threats. “I stirred it up. Do I regret it? Who knows?” he said when we met in Australia. “It’s caused a lot of chaos.”
But, he added, “this is the bed that I’ve laid.” When we spoke again, months later, he backtracked, saying he “got sucked into” defending THORChain in 2024 and 2025 because he was involved from 2018 to 2021 and has “a perspective on how the protocol operates.”
Centralized?
Ryan Treat, a retired US Army veteran, woke up one morning in January 2025 to some disturbing activity on X. “My heart sank,” he says. THORFi, the THORChain program he’d used to earn interest on the bitcoin he’d planned to save for his retirement, had frozen all accounts—but that didn’t make sense.
THORFi featured a lending and saving program said to give users “complete control” and self-custody of their crypto, meaning they could withdraw it at any time.
Treat was no crypto amateur. He bought his first bitcoin at around “$5 apiece,” he says, and had always kept it off centralized exchanges that would maintain custody of his wallets. He liked THORChain because it claimed to be decentralized and permissionless. “I got into bitcoin because I wanted to have government-less money,” he says.
We were told it was decentralized. Then you wake up one morning and read this guy had an admin mimir.
Many who’d used THORFi lending and saving programs felt similarly. Users I interviewed differentiated THORChain from centralized lending platforms like BlockFi and Celsius, both of which offered extraordinarily high yields before filing for bankruptcy in 2022. “I viewed THORChain as a decentralized system where it was safer,” says Halsey Richartz, a Florida-based THORFi creditor, with “vanilla, 1% passive yield.” Indeed, users I spoke with hadn’t felt the need to monitor their THORFi deposits. “Only your key can be used to withdraw your funds,” the product’s marketing materials insisted. “Savers can withdraw their position to native assets at any time.”
So on January 9, when the “leena” account announced that an admin key had been used to pause withdrawals, it took THORFi users by surprise—and seemed to contradict the marketing messaging around decentralization. “We were told that it was decentralized, and you wake up one morning and read an article that says ‘This guy, JP, had an admin mimir,’” says Treat, referring to Thorbjornsen, “and I’m like, ‘What the fuck is an admin mimir?’”
The admin mimir was one of “a bunch of hard-coded admin keys built into the base code of the system,” says Jonathan Reiter, CEO of the blockchain intelligence company ChainArgos. Those with access to the keys had the ability to make executive decisions on the blockchain—a function many THORChain users didn’t realize could supersede the more democratic decisions made by node votes. These keys had been in THORChain’s code for years and “let you control just about anything,” Reiter adds, including the decision to pause the network during the hacks in 2021 that resulted in a loss of more than $16 million in assets.
Thorbjornsen says that one key was given to Nine Realms, while another was “shared around the original team.” He told me at least three people had them, adding, “I can neither confirm nor deny having access to that mimir key, because there’s no on-chain registry of the keys.”
Regardless of who had access, Thorbjornsen maintains that the admin mimir mechanism was “widely known within the community, and heavily used throughout THORChain’s history” and that any action taken using the keys “could be largely overruled by the nodes.” Indeed, nodes voted to open withdrawals back up shortly after the admin key was used to pause them. By then, those burned by THORFi argue, the damage had already been done. The executive pause to withdrawals, for some, signaled that something was amiss with THORFi. This led to a bank run after the pause was lifted, until the nodes voted to freeze withdrawals permanently (which Thorbjornsen had suggested in a since-deleted post on X), separating users from crypto worth around $200 million in US dollars on January 23. THORFi users were then offered a token called TCY (THORChain Yield), which they could claim with the idea that, when its price rose to $1, they would be made whole. (The price, as of writing, sits at $0.16.)
Who used the key? Thorbjornsen maintains he didn’t do it, but he claims he knows who did and won’t say. He says he’d handed over the “leena” account and doesn’t “have access to any of the core components of the system,” nor has he for “at least three years.” He implies that whoever controlled “leena” at the time used the admin key to pause network withdrawals.
A video released by Nine Realms on February 20, 2025, names Thorbjornsen as the activator of the key, stating, “JP ended up pausing lenders and savers, preventing withdrawals so that we can work out … [a] payback plan on them.” Thorbjornsen told me the video was “not factual.”
Multiple blockchain analysts told me it would be extremely difficult to determine who used the admin mimir key. A month after it was used to pause the network, THORChain said the key had been “removed from the network.” At least you can’t find the words “admin mimir” in THORChain’s base code; I’ve looked.
Culpability
After the debacle of the THORFi withdrawal freeze, Richartz says, he tried to file reports with the Miami-Dade Police Department, the Florida Department of Law Enforcement, the FBI, the Securities and Exchange Commission, the Commodity Futures Trading Commission, the Federal Trade Commission, and Interpol. When we spoke in November, he still hadn’t been able to file with the city of Miami. They told him to try small claims court.
“I was like, no, you don’t understand … a post office box in Switzerland is the company address,” he says. “It underscored to me how little law enforcement even knows about these crimes.”
As for the Bybit hack, at least one government has retaliated against those who facilitate blockchain projects. Last April German authorities shut down eXch, an exchange suspected of using THORChain to process funds Lazarus stole from Bybit, says Julia Gottesman, cofounder and head of investigations at the cybersecurity group zeroShadow. Australia, she adds, where Thorbjornsen was based, has been “slow to try to engage with the crypto community, or any regulations.”
KAGAN MCLEOD
In response to requests for comment, Australia’s Department of Home Affairs wrote that at the end of March 2026, the country’s regulatory powers will expand to include “exchanges between the same type of cryptocurrency and transfers between different types.” They did not comment on specific investigations.
Crypto and finance experts disagree about whether THORChain engaged in money laundering, defined by the UN as “the processing of criminal proceeds to disguise their illegal origin.” But some think it fits the definition.
Shlomit Wagman, a Harvard fellow and former head of Israel’s anti-money-laundering agency and its delegation to the Financial Action Task Force (FATF), thinks the Bybit activity was money laundering because THORChain helped the hackers “transfer the funds in an unsupervised manner, completely outside of the scope of regulated or supervised activity.”
And by helping with conversions, Carlsen says, THORChain enabled bad actors to turn stolen crypto into usable currency. “People like [Thorbjornsen] have a personal degree of culpability in sustaining the North Korean government,” he says. Thorbjornsen counters that THORChain is “open-source infrastructure.”
Meanwhile, just days after the hack, Bybit issued a 10% bounty on any funds recovered. As of mid-January this year, between $100 million and $500 million worth of those funds in US dollars remain unaccounted for, according to Gottesman of zeroShadow, which was hired by Bybit to recover funds following the hack.
Thorbjornsen hacked
For Thorbjornsen, it’s just another day at the casino. That’s the comparison he made during his regrettable 7 News Spotlight interview about the Bybit heist, and he repeated it when we met. “You go to a casino, you play a few games, you expect to lose,” he told me. “When you do actually go to zero, don’t cry.”
Thorbjornsen, it should be noted, has lost at the casino himself.
In September, he says, he got a Telegram message from a friend, inviting him to a Zoom meeting. He accepted and participated in a call with people who had “American voices.”
Ultimately, Thorbjornsen describes himself as a guy who’s had a bad year, fending off “threat vectors” left and right.
After the meeting, Thorbjornsen learned that his friend’s Telegram had been hacked. Whoever was responsible had used the Zoom link to remotely install software on Thorbjornsen’s computer, which “got access to everything”—his email, his crypto wallets, a bitcoin-based retirement fund. It cost him at least $1.2 million. The blockchain sleuth known as ZachXBT traced the funds and attributed the hack to North Korea.
ZachXBT called it “poetic.”
Ultimately, Thorbjornsen describes himself as a guy who’s had a bad year. He says he had to liquidate his crypto assets because he’s dealing with the fallout of a recent divorce. He also feels he is fending off “threat vectors” left and right. More than once, he asked if I was a private investigator masquerading as a journalist.
Still, his many contradictions don’t inspire confidence. He doesn’t have any more crypto assets, he says. However, the crypto wallet he shared with me so I could pay him back for lunch showed that it contained assets worth more than $143,000 in US dollars. He now says it wasn’t his wallet. He says he doesn’t control THORChain’s social media, but he’d also told me that he runs the @THORChain X account (later backtracking and saying the account is “delegated” to him for trickier questions).
He insists that he does not care about money. He says that in the robot future, the AI-powered hive mind will become our benevolent overlord, rendering money obsolete, so why give it much thought? Yet as we flew back from the vineyard, he pointed out his new house from the helicopter. It resembles a compound. He says he lives there with his new wife.
Multiple people I spoke with about Thorbjornsen before I met him warned me he wasn’t trustworthy, and he’s undeniably made fishy statements. For instance, the presence of a North Korean flag in a row of decals on the tail of his helicopter suggested an affinity with the country for which THORChain has processed so much crypto. Thorbjornsen insists he had requested the flag of Australia’s Norfolk Island, calling the mix-up “a complete coincidence.” The flags were gone by the time of our flight, apparently removed during a recent repair.
“Money is a meme,” he says. “Money does not exist.” Meme or not, it’s had real repercussions for those who have interacted with THORChain, and those who wound up losing have been looking for someone to blame.
When I spoke with Thorbjornsen again in January, he appeared increasingly concerned that he is that someone. He’s spending more time in Singapore, he told me. Singapore happens to have historically denied extraditions to the US for money-laundering prosecutions.
Jessica Klein is a Philadelphia-based freelance journalist covering intimate partner violence, cryptocurrency, and other topics.
LinkedIn Premium has seen a 50% rise in subscriptions over the past two years.
Meta also shut down the separate Messenger app for Windows and Mac last year.
Reddit is already seeing an influx of interest in the event.
The platform’s add-on subscription has become a valuable revenue driver for the business.