-
North Korea’s State-Sponsored Cyber Theft: How a Country Funds Itself Through Hacking
On February 21, 2025, the CEO of Bybit—a Dubai-based cryptocurrency exchange—approved what appeared to be a routine transaction. The user interface showed the correct destination address. The multi-signature security system required multiple executives to sign off, and they did. The transaction looked legitimate at every layer of verification a human being could perform. It wasn’t. North Korea’s Lazarus Group had compromised the interface of Safe{Wallet}, a third-party wallet tool that Bybit used for transfers between cold storage and hot wallets. The interface displayed one address. The code sent funds to another. By the time anyone noticed, 400,000 Ethereum—worth approximately $1.5 billion—had been transferred to wallets controlled by Pyongyang’s military intelligence apparatus. It was the largest cryptocurrency theft in history, executed through a fake button on a screen.
Within 48 hours, at least $160 million had been laundered. By March 20—less than a month later—Bybit’s CEO confirmed that attackers had converted 86 percent of the stolen Ethereum to Bitcoin. The money was gone, distributed across a laundering infrastructure that blockchain analysts describe as industrialized, following a structured 45-day pipeline from theft to usable currency. According to Chainalysis’s Crypto Crime Report, North Korean hackers stole $2.02 billion in cryptocurrency in 2025 alone—a 51 percent increase over the $1.3 billion stolen in 2024. The cumulative total since 2017 exceeds $6.75 billion. United Nations monitors estimate that cryptocurrency theft now constitutes approximately 13 percent of North Korea’s GDP.
This is not a criminal enterprise. This is a national economy.
How a country became a hacking operation
The Lazarus Group is affiliated with North Korea’s Reconnaissance General Bureau—the regime’s primary intelligence agency. According to a North Korean defector, the unit is known internally as the 414 Liaison Office. It first gained international attention in 2014 by destroying Sony Pictures’ network infrastructure in retaliation for The Interview, a film depicting the assassination of Kim Jong-un. The hackers deployed wiper malware that erased data across Sony’s systems while publicly leaking internal communications—a political operation, not a financial one.
The pivot to financial crime came in 2016 with the Bangladesh Bank heist. Lazarus issued 35 fraudulent instructions through the SWIFT international banking network to transfer nearly $1 billion from the Federal Reserve Bank of New York’s account belonging to Bangladesh’s central bank. Thirty of the transactions were blocked when a misspelled word in one instruction triggered a review. Five got through. The group escaped with $81 million—a figure that, by current standards, would be a slow Tuesday.
The cryptocurrency era transformed the operation’s scale. Traditional banking systems have compliance departments, transaction limits, correspondent bank oversight, and regulatory checkpoints. Cryptocurrency has smart contracts, multi-signature wallets, and decentralized exchanges with varying levels of security, operated by companies headquartered in jurisdictions with inconsistent enforcement. For a nation-state hacking operation, the cryptocurrency ecosystem is a softer target than the SWIFT network by orders of magnitude.
The progression since 2017: Banco del Austro in Ecuador ($12 million), Vietnam’s Tien Phong Bank ($1 million), Taiwan’s Far Eastern International Bank ($60 million), then the escalation into crypto—KuCoin ($275 million in 2020), the Ronin Network powering Axie Infinity ($625 million in 2022), Atomic Wallet ($100 million in 2023), WazirX in India ($235 million in 2024), and then Bybit ($1.5 billion in February 2025). The trajectory is exponential, and the operational tempo is accelerating: by mid-2025, Lazarus was executing major heists roughly every 20 days.
How they actually get in
The Lazarus Group’s primary weapon is not technical sophistication. It’s patience. Their attack methodology targets humans, not code.
The Ronin Network hack—$625 million—started with a fake job offer. A Lazarus operative, posing as a recruiter, contacted an engineer at Sky Mavis (the company behind Axie Infinity) through LinkedIn with a fabricated employment opportunity. The engineer downloaded a document that contained malware. That single compromised machine gave the attackers access to the validator nodes that secured the Ronin bridge, and from there, access to the funds.
The Bybit hack started with a compromised developer laptop. On February 4, 2025, a developer at Safe{Wallet} received what appeared to be a routine request. Their Apple MacBook became the entry point. Within 17 days, the attackers had manipulated the wallet’s front-end interface to redirect a legitimate-looking transaction. The multi-signature security system—designed specifically to prevent single-point-of-failure theft—approved the fraudulent transfer because the fraud existed at the visual layer, not the cryptographic layer. The keys were valid. The signatures were authentic. The destination was wrong.
Beyond direct hacking, North Korea has deployed what researchers call the “Wagemole” strategy—embedding covert IT workers inside legitimate companies worldwide. Operatives obtain remote technical positions using fraudulent identities or through front companies, function as normal employees while providing intelligence to hacking teams, and in some cases directly facilitate theft by providing credentials or disabling security systems. In 2024 alone, more than a dozen cryptocurrency companies were infiltrated by North Korean operatives posing as IT contractors. A Maryland man was sentenced in December 2025 to 15 months in prison for allowing North Korean nationals in Shenyang, China, to use his identity for employment at U.S. companies—including a contract at the Federal Aviation Administration. He was paid over $970,000 for software development work performed by overseas conspirators.
Why they can’t be stopped (yet)
North Korea has no extradition treaties. No Interpol cooperation. No financial system to freeze. The Lazarus Group operates from Pyongyang with functional impunity. The U.S. Treasury has sanctioned over 100 Lazarus-linked wallet addresses, but the group creates new ones. The FBI issues arrest warrants for operators it will never arrest. International sanctions on North Korea are among the most comprehensive ever imposed—and cryptocurrency theft is the mechanism by which the regime circumvents them.
The laundering infrastructure is equally resilient. Chainalysis analysis reveals a structured, multi-wave pipeline: within hours of a theft, stolen funds begin moving through DeFi protocols and mixing services. Funds from separate heists are blended together—money from Stake.com ends up in wallet addresses used for Atomic Wallet laundering, CoinEx proceeds flow through addresses tied to previous operations. This intentional commingling creates noise that makes individual theft attribution nearly impossible. Analysts can trace fragments, but only about 15 percent of stolen funds are ever recovered. The stolen cryptocurrency is converted to Bitcoin (highest liquidity, global acceptance, resistance to devaluation), moved through privacy-enhancing mixers, and eventually converted to fiat currency through intermediaries in jurisdictions with weak enforcement.
The Center for Strategic and International Studies calls this “cyber-enabled state terrorism.” The label is accurate. Every dollar stolen funds North Korea’s nuclear weapons and ballistic missile programs. Every Bitcoin heist buys missile fuel. The February 2025 Bybit theft alone—$1.5 billion—exceeded the entire annual GDP of several sovereign nations, extracted through a single manipulated interface in 81 seconds of approved transactions.
What it means for the concept of a heist
North Korea’s cryptocurrency operation redefines what a heist is. The Gardner Museum theft required two men in police uniforms, 81 minutes inside a building, and the physical removal of 13 canvases. The Bybit theft required a compromised laptop, a manipulated interface, and a CEO clicking a button he’d been designed to trust. The Gardner paintings are worth $500 million and are unsellable. The Bybit Ethereum was worth $1.5 billion and was 86 percent laundered within a month.
The traditional heist is constrained by physical access, physical removal, and physical fencing of stolen goods. The North Korean model removes all three constraints. Access is digital. Removal is instantaneous. And cryptocurrency—unlike a Vermeer—can be laundered into fungible currency through automated infrastructure that operates 24 hours a day across every jurisdiction on earth. The heist of the century is no longer a once-in-a-generation event. It’s a quarterly revenue target for a nuclear-armed state that has turned theft into a line item on its national budget.
We cover North Korea’s cyber operations alongside the Gardner Museum theft, the economics of stolen property, and the full history of audacious theft across our Greatest Heists course—including why the most successful heist crew in history doesn’t wear masks, carry guns, or leave the building. They sit at keyboards in Pyongyang and steal more in an afternoon than most bank robbers dream of in a lifetime.
-
The Hum: The Unexplained Low-Frequency Sound That Drives People Crazy in Dozens of Cities Worldwide
In the early 1990s, residents of Taos, New Mexico, started complaining about a low-frequency humming sound that wouldn’t stop. It was there when they went to bed and there when they woke up—a steady, throbbing drone, like a diesel engine idling somewhere over the horizon. It was louder at night, louder indoors, and impossible to locate. Not everyone could hear it. Roughly 2 percent of the population reported the sound. The other 98 percent heard nothing. The complaints were persistent enough that Congress funded an investigation. A team from Los Alamos National Laboratory, Sandia National Laboratories, and the University of New Mexico deployed specialized acoustic equipment tuned to frequencies between 8 and 80 hertz—the range where sound registers more as vibration than tone. They found that the hearers were telling the truth: something was being perceived, each person at a slightly different frequency between 32 and 80 hertz. They could not identify a source. The investigation ended inconclusively. The sound did not.
The Taos Hum was not the first and was nowhere close to the last. Bristol, England, reported a persistent thrumming in the 1970s—about 800 people heard it. It was tentatively blamed on vehicular traffic and factories running 24-hour shifts, but never definitively explained, and the reports eventually faded. Windsor, Ontario, erupted in late 2011 with a low droning vibration loud enough to provoke 22,000 reports to officials in a single evening in 2012. Kokomo, Indiana. Largs, Scotland. Auckland, New Zealand. Bondi, Australia. Frankfurt and Darmstadt, Germany. San Francisco‘s Sunset District, where residents reported it as recently as 2024. The Hum has been documented on every inhabited continent, and the case files share the same strange profile: a low-frequency drone, typically between 30 and 80 hertz, heard indoors more than outdoors, worse at night, worse in quiet environments, perceived by a small minority of the population while the majority hears nothing at all.
What the investigations found
Some Hums have been solved. The Windsor Hum was traced, with reasonable confidence, to Zug Island—a heavily industrialized section of River Rouge, Michigan, across the Detroit River from Windsor. Canadian officials identified the area as the likely source, but jurisdictional politics complicated the investigation: local authorities couldn’t access the island, and U.S. Steel, which operated a steel mill there, said no new equipment had been installed around the time the noise became noticeable. The resolution came accidentally. When the blast furnaces were deactivated in April 2020, during the pandemic shutdowns, the Hum stopped. When operations resumed, the Hum returned. In Darmstadt, Germany, investigators in 2022 identified multiple sources: two faulty air conditioner units, a faulty heat pump, and three structural noise protection measures on energy generation plants that were themselves producing low-frequency noise. In Kokomo, industrial fans were implicated, though some reports persisted after the fans were addressed.
These solved cases share a common mechanism: industrial equipment generating low-frequency noise that propagates through the ground or air and is amplified by the resonant properties of certain buildings. A room with the right dimensions can amplify a faint 40-hertz signal into something perceptible, the way a wine glass vibrates when you hit the right frequency. Low-frequency sound penetrates walls more effectively than higher frequencies—bass travels through structures that block treble—which explains why the Hum is louder indoors. It’s louder at night because ambient noise drops, unmasking sounds that were always present but drowned out during the day. It’s louder in suburban and rural environments than in cities for the same reason: less background noise.
But the industrial explanation doesn’t account for all the cases. The Taos Hum investigation found no industrial source. The Bristol Hum was never definitively explained. Auckland researchers found some low-frequency sources, silenced them, and the complaints continued. The Hum in Kerry County, Ireland, was investigated and remains unexplained. The pattern—some cases explained by identifiable mechanical sources, others remaining stubbornly unresolved—suggests that “the Hum” is not a single phenomenon. It’s a symptom that can have multiple causes, some of which are industrial, some of which may be biological, and some of which haven’t been identified.
The biology of hearing things that aren’t there (or are)
The human ear is not a passive microphone. It generates its own sounds—called spontaneous otoacoustic emissions—produced by the motion of the outer hair cells in the cochlea. Studies show that 38 to 60 percent of adults with normal hearing produce these emissions, though most people are unaware of them. In quiet environments, some individuals perceive their own otoacoustic emissions as a faint hissing, buzzing, or humming. The Taos investigation considered this as a possible explanation: the Hum might not be coming from outside the ear but from inside it.
This hypothesis explains some features of the phenomenon—why only a small percentage of people hear it, why it’s worse in quiet environments, why earplugs sometimes make it louder rather than softer (blocking external noise unmasks the internal signal)—but it doesn’t explain the geographic clustering. If the Hum were purely a biological artifact, it should be distributed randomly across the population, not concentrated in specific towns during specific time periods. The geographic pattern suggests an external stimulus, even if the perception of that stimulus is mediated by individual differences in auditory sensitivity.
Low-frequency tinnitus is another biological candidate. Tinnitus—the perception of sound without an external source—typically manifests as high-pitched ringing, but a subset of cases involve low-frequency perception in the range of the Hum. Some researchers have proposed that the Hum represents a form of tinnitus that is triggered or modulated by environmental low-frequency noise too faint for most people to perceive but sufficient to activate auditory responses in sensitized individuals. Under this model, the industrial source doesn’t have to be loud enough for most people to hear. It just has to be present enough to trigger a disproportionate perceptual response in the 2 percent of the population whose auditory systems are tuned to those frequencies.
A 1973 university study of 50 Hum complainants found the sound always peaked between 30 and 40 hertz, was heard only during cool weather with a light breeze, and was more common in early morning. Philip Dickinson suggested at an Institute of Biology conference that year that the sound could result from the jet stream shearing against slower-moving air, possibly amplified by power line structures or by rooms with corresponding resonant frequencies. Another acoustics researcher dismissed this as “absolute nonsense.” The disagreement is characteristic of the field: every proposed explanation accounts for some features of the data while failing to explain others, and the researchers who study the Hum spend as much time arguing with each other as with the phenomenon.
Why it matters beyond the sound
The Hum has driven at least one person to suicide in England. Others report chronic insomnia, headaches, nausea, nosebleeds, and diarrhea. In Largs, Scotland, residents moved away. In Windsor, the 22,000 reports to officials in a single night reflected a community that had been sleep-deprived and frustrated for months. The Hum is not a curiosity for the people who hear it. It’s a quality-of-life crisis that they often can’t prove to their neighbors, their doctors, or their local government—because the person standing next to them in the same room, at the same time, hears nothing.
This is what makes the Hum a genuinely interesting epistemological problem rather than just an acoustic one. It exists at the intersection of physics, biology, psychology, and infrastructure—a sound that may be real, may be internal, may be both, and whose investigation requires expertise in acoustics, otology, environmental engineering, and psychophysics, all operating simultaneously. The solved cases prove that external low-frequency sources exist and can cause the reported symptoms. The unsolved cases prove that the solved explanations don’t cover everything. The biological evidence proves that the human ear can generate perceptions that have no external correlate. And the geographic clustering proves that biology alone doesn’t explain the pattern.
The Hum is, in some respects, a perfect Fortean phenomenon: real enough to investigate, elusive enough to resist explanation, distributed widely enough to suggest a systematic cause, and variable enough to prevent any single theory from closing the case. Charles Fort would have collected the reports, filed them, and waited. Half a century of acoustic science has, essentially, done the same thing.
We cover the Hum alongside ball lightning, UAP encounters, and the full taxonomy of phenomena that resist clean scientific explanation across our Fortean Phenomena course—including why the most maddening sound in the world is one that only 2 percent of people can hear.
-
Japan’s Elder Care Robots: What Happens When a Country Builds Robots for Its Aging Population
Tokyo’s Shin-tomi nursing home uses 20 different robot models to care for its residents. PARO, a fluffy animatronic harp seal that took over a decade to develop and received roughly $20 million in government funding, responds to touch and speech by moving its head, blinking, and playing recordings of seal cries. SoftBank’s Pepper leads afternoon exercise sessions and runs scripted dialogues. Tree, a walking rehabilitation device, crawls along the floor and tells patients where to place their next step in a gentle feminine voice—”right, left, well done!” Panasonic’s robotic bed transforms into a wheelchair. Monitoring systems track falls. The facility has become a showcase that more than 100 foreign delegations visited in a single year, from China, South Korea, the Netherlands, and elsewhere—countries watching Japan navigate a demographic crisis they know is heading their way.
Japan has 36.25 million people aged 65 or older as of 2024, roughly 29 percent of the population. By 2065, one in every 2.6 people in Japan will be 65 or older. The country has the highest life expectancy and the largest proportional elderly population of any nation on earth. The birth rate has been declining for decades. The labor shortage in elder care is acute: as of recent data, there is only one applicant for every 4.25 job openings in the sector. The Ministry of Health, Labour and Welfare projected a shortage of 370,000 caregivers by 2025. Projections for 2040 suggest a shortfall of 11 million workers across all sectors. Japan isn’t building care robots because robots are cool. Japan is building care robots because the math doesn’t work any other way.
What the robots actually do (and don’t do)
The most detailed ethnographic account of care robots in practice comes from a researcher who spent over 18 months in Japanese elder care facilities, including extended observation at a home trialing three robots: Hug (a lifting device), PARO (the seal), and Pepper (the humanoid). The findings are instructive for what they reveal about the distance between the technology’s promise and its operational reality.
Hug, the lifting robot designed to prevent care workers from manually lifting residents, was abandoned within days. Staff found it cumbersome and time-consuming to wheel from room to room—the time spent maneuvering the device cut into the time they had available to actually interact with residents. The robot was solving a physical problem (back strain from lifting) while creating a logistical one (reduced care time per resident), and the staff decided the tradeoff wasn’t worth it.
PARO was received more favorably. Residents responded to the soft, reactive seal in ways that suggested genuine emotional engagement. But complications emerged quickly. One resident kept trying to “skin” PARO by pulling off its synthetic fur. Another developed such an intense attachment that she refused to eat meals or go to bed without the robot by her side. Staff ended up monitoring PARO’s interactions closely rather than being freed from monitoring duties—the opposite of the intended labor-saving effect. PARO didn’t reduce repetitive behavior patterns in residents with severe dementia, which had been one of the primary hoped-for outcomes. And because PARO can’t move independently, staff had to carry it from room to room, adding a task rather than eliminating one.
Pepper’s deployment followed a similar pattern of expectation meeting friction. Instead of freeing a care worker from leading afternoon recreation sessions, Pepper required a care worker to spend time booting it up, wheeling it into position, and managing the session alongside it. Staff found Pepper difficult to set up. It couldn’t respond to voice commands or move independently—functions that SoftBank acknowledged were needed but hadn’t yet been implemented. Pepper with the Care Prevention Gymnastics Exercises application could facilitate a 40-minute exercise program, but a care worker still had to be present throughout.
The pattern across all three robots is consistent: each one solved a narrow technical problem while creating new operational burdens that partially or fully offset the labor savings. The lifting robot was too cumbersome. The therapeutic seal required more supervision, not less. The humanoid exercise leader needed a human assistant.
Why this keeps happening
The disconnect between care robot demonstrations and care robot reality has a structural explanation. Care is not a logistics problem. It’s a relational activity that happens between people, and the parts of care that are most labor-intensive—emotional engagement, judgment under uncertainty, adapting to a resident’s changing mood and condition—are precisely the parts that robots are worst at.
The tasks robots handle well are the ones that were never the primary bottleneck: leading a scripted exercise routine, playing pre-recorded sounds, transforming a bed into a wheelchair. The tasks that consume the most caregiver time and cause the most burnout—managing behavioral complications of dementia, providing emotional support to residents who are frightened or confused, making real-time clinical judgments about a resident’s condition—require exactly the kind of contextual, adaptive, emotionally intelligent interaction that current robotics can’t deliver.
Some researchers have raised ethical concerns that cut deeper than efficacy. Using toy-like robots with dementia patients raises questions about infantilization—treating cognitively impaired adults as if they were children comforted by stuffed animals. PARO’s therapeutic seal form is specifically designed to trigger nurturing responses, but residents with cognitive impairment may believe it’s a real animal, raising the question of whether therapeutic benefit achieved through deception is acceptable. Care workers in multiple studies have expressed discomfort with this dynamic. Others have raised privacy concerns about robots with cameras and microphones operating in residents’ living spaces, with some staff reporting the feeling that the robots were monitoring their work.
The demographic argument doesn’t go away
None of these complications change the underlying math. Japan will have one person aged 65 or older for every 2.6 people in the total population by 2065. The caregiver shortage isn’t projected to close through immigration—as of 2017, only 18 foreigners held Japan’s nursing care visa. The Specified Skilled Worker System introduced in 2019 has expanded foreign labor in care, but the numbers remain far short of the projected deficit. The healthcare and welfare sector is on track to become the largest industry in Japan, and there aren’t enough humans to staff it.
The Ministry of Economy, Trade and Industry estimates the domestic care robot industry will reach ¥400 billion (roughly $2.7 billion) by 2035, when a third of Japan’s population will be 65 or older. The global market in 2016 was $19.2 million. The growth curve is steep because the need is not optional—it’s demographic arithmetic.
The robots that are gaining traction are not the charismatic humanoids that generate media coverage. They’re the unglamorous operational tools: monitoring systems that detect falls and irregular behavior, robotic beds that reduce transfer injuries, sensor networks that track resident movement patterns and alert staff to anomalies, telepresence systems that allow remote caregivers to check on residents without being physically present. These don’t make for compelling photographs, but they address real workflow bottlenecks without requiring residents to form emotional relationships with machines.
What Japan is actually teaching the world
Japan’s two decades of investment in elder care robots have produced a result more valuable than any individual robot: a detailed, field-tested body of evidence about what happens when you deploy technology into a care environment. The lessons are consistent and transferable. Robots that create operational burdens for staff get abandoned regardless of their technical capabilities. Robots that require human supervision don’t reduce labor costs. Therapeutic robots raise ethical questions that the engineering alone can’t answer. And the tasks most in need of automation—the emotionally complex, contextually adaptive, judgment-intensive interactions that define good care—remain beyond current robotic capability.
The countries sending delegations to Shin-tomi aren’t just shopping for robots. They’re studying what happens when a technologically advanced society confronts an irreversible demographic shift and discovers that the hardest problems in elder care aren’t the ones that engineering solves most easily. Germany, China, Italy, South Korea, and eventually the United States will face the same mathematics. Japan got there first—and what it found is that the robots help, but they don’t fix, because what’s breaking isn’t a machine. It’s a labor market, a social contract, and a set of political choices about who takes care of the people who can no longer take care of themselves.
We cover Japan’s elder care robotics alongside the uncanny valley, the humanoid robot race, and the full landscape of human-robot interaction across our Humanoid Robots & Drones course—including why the most expensive robot in a nursing home is often the one nobody uses.
-
Programmable Matter: Materials That Change Shape on Command and Why They’re Still in the Lab
In 2002, Seth Goldstein at Carnegie Mellon University coined the term “claytronics” to describe a material that doesn’t exist yet: a substance made of millions of tiny robots called catoms—claytronic atoms—each a few microns in diameter, capable of computing, communicating via electrostatics, clinging to neighboring catoms, transferring energy between them, and rearranging themselves into any three-dimensional shape on command. Scoop up a lump of this material, tell it what to become, and it becomes it—a wrench, a phone, a replacement part for a machine you’ve never seen. When you’re done, it dissolves back into formless goo, waiting for the next instruction.
That was 24 years ago. The catoms don’t exist. The goo doesn’t exist. What exists, in 2026, is a collection of genuinely impressive but fundamentally limited demonstrations across several related-but-distinct fields, none of which are close to the vision that Goldstein described and every science fiction franchise since T-1000 has promised. Programmable matter remains one of the most compelling ideas in materials science and one of the most stubborn gaps between concept and execution.
What “programmable matter” actually means
The term covers at least three distinct approaches that share a name but not much else.
The first is claytronics proper—Goldstein’s original vision of modular robotic matter. Tiny robots, each containing a processor, power system, communication hardware, and actuators, self-assemble into macroscopic shapes. Researchers have demonstrated small-scale 2D self-assembly with centimeter-scale modules. Going from centimeter-scale 2D formations to micron-scale 3D programmable matter requires solving power delivery, communication bandwidth, actuation precision, thermal management, and manufacturing scalability simultaneously, at a scale where the individual units are smaller than the width of a human hair. Hod Lipson at Cornell has said that modular robotic systems will eventually be “the only way to ensure sustainable production”—in the very long term, meaning centuries.
The second is 4D printing—3D-printed objects made from stimuli-responsive materials that change shape after fabrication in response to external triggers like heat, water, light, pH, or magnetic fields. Skylar Tibbits at MIT introduced the term during a 2013 TED talk. The “fourth dimension” is time: the printed object transforms into its target shape after leaving the printer. Shape memory polymers are the most studied material class—they can be deformed into a temporary shape and then return to their original geometry when heated past a transition temperature. Liquid crystal elastomers change shape in response to light. Hydrogels swell or contract in response to moisture. Researchers have 4D-printed self-folding origami structures, biodegradable soft robots, biomedical scaffolds, and ceramics that deform during pyrolysis.
4D printing is real, published, and accelerating in the research literature. It is also, by Goldstein’s definition, not truly programmable—it’s responsive. The material responds to a stimulus according to properties locked in during fabrication. A 4D-printed structure heated in a water bath will always fold in the same direction because its geometry and material composition determine the fold. It doesn’t run code. It doesn’t make decisions. It executes a pre-programmed physical transformation, which is a meaningful distinction from a material that can adopt arbitrary shapes on demand.
The third is programmable mechanical metamaterials—engineered structures with unit cells whose mechanical properties (stiffness, damping, Poisson’s ratio, energy absorption) can be switched between discrete states. Electromagnetic metamaterials with unit cells in binary “0” or “1” states, controlled by diodes and field-programmable gate arrays, can steer electromagnetic waves programmatically. Mechanical versions use bistable elements, shape memory alloys, or active hinges to switch between rigid and flexible states. These are genuinely programmable in the computational sense—you can reprogram the structure’s behavior without rebuilding it—but they’re metamaterials, not matter. They’re engineered architectures, typically centimeter-scale or larger, not materials you could scoop up and reshape.
Why catoms don’t exist
The core problem is that making a functional robot at the micron scale is orders of magnitude harder than making one at the centimeter scale, and every system that a catom needs—computation, communication, actuation, power—has a minimum viable size that current fabrication technology can’t simultaneously achieve at the dimensions required.
MEMS (micro-electromechanical systems) technology can build individual components at the micron scale. What it can’t yet do is integrate all the necessary subsystems—a processor capable of running coordination algorithms, an energy harvesting or storage system, electrostatic adhesion actuators, and communication hardware—into a single package small enough that millions of them could approximate a continuous material. The individual technologies exist. The integration doesn’t. And even if fabrication were solved, the coordination problem—getting millions of autonomous units to compute their target positions, negotiate movement sequences with neighbors, avoid deadlocks and collisions, and converge on a stable macroscopic shape—is a distributed computing challenge that gets harder, not easier, as the number of units increases.
Power is particularly stubborn. A micron-scale robot needs energy to compute, communicate, actuate, and maintain adhesion. If it harvests energy from its environment (light, vibration, chemical gradients), the harvesting mechanism needs physical area, which competes with the other subsystems for space on an already impossibly small device. If it receives energy from neighbors via electrostatic transfer, the transfer efficiency and the maximum power delivery rate constrain what the catom can do and how fast it can do it. Every catom in the interior of a formation—surrounded on all sides by other catoms—faces the additional problem of receiving power from neighbors who are themselves receiving power from their neighbors, creating cascading efficiency losses.
What’s actually advancing
The honest state of the field in 2026 is that 4D printing is a legitimate and growing research area producing functional demonstrations, while claytronics-style programmable matter remains a theoretical goal with no near-term path to realization.
4D printing publications have grown rapidly since 2013. Researchers have demonstrated shape memory polymer composites that fold into complex origami geometries, ceramic structures that deform during processing, hydrogel-based soft robots that move without motors or electronics, food packaging that changes color to indicate spoilage, and biomedical implants that reshape themselves after insertion. Market analyses project significant growth in the commercial 4D printing sector through 2029, driven by applications in construction, healthcare, military, automotive, and textiles—though the still-high cost of 4D printers is identified as the major brake on market expansion.
Programmable mechanical metamaterials are producing results in vibration isolation, energy absorption, and shape morphing for aerospace and soft robotics applications. These structures can switch between configurations—stiff to flexible, positive to negative Poisson’s ratio—in response to magnetic fields, temperature changes, or mechanical inputs. They are programmable in a meaningful sense, but they are discrete engineered structures, not continuous materials.
Synthetic biology represents a parallel approach that sidesteps the robotic fabrication problem entirely. Autodesk’s research head Gord Kurtenbach has argued that biological systems are already fully programmable materials—DNA encodes arbitrary structural information, cells self-assemble into complex architectures, and organisms grow and adapt without external motors or electronics. The question is whether synthetic biology can produce materials that respond to commands rather than evolutionary pressures. If a structure could be grown from a “seed” and then programmed to adjust as required, that would be a fundamentally different paradigm from either 4D printing (which has finality—the transformation is predetermined) or claytronics (which requires solving the robot miniaturization problem).
The gap
The distance between where programmable matter is and where the concept promises it could be is one of the largest in materials science. The vision—arbitrary shape-shifting on command, matter as a general-purpose substrate—requires either solving micron-scale robotic integration (claytronics), inventing materials that compute (something between 4D printing and synthetic biology), or discovering a mechanism that nobody has proposed yet. The timeline for any of these is not years. It might be decades. It might, as Lipson suggested, be centuries.
What exists right now is a constellation of related technologies—4D printing, metamaterials, shape memory alloys, responsive hydrogels, modular robotics—each of which captures a fragment of the programmable matter vision without achieving the whole. They’re real, they’re useful, and they’re advancing. They’re also not catoms. The lump of material that becomes whatever you need it to be, and then becomes something else, remains where it has been since 2002: in the concept paper, not in the lab.
We cover programmable matter alongside synthetic biology, quantum computing, and the full landscape of technologies that could reshape civilization across our Moonshot 2169 course—including why the most honest thing you can say about a material that changes shape on command is that the command still doesn’t exist.
-
Space-Based Solar Power in 2026: Beaming Energy from Orbit to Earth
On May 22, 2023, a receiver on the roof of the Gordon and Betty Moore Laboratory of Engineering at Caltech’s campus in Pasadena detected a microwave signal beamed from orbit. The signal came from MAPLE—Microwave Array for Power-transfer Low-orbit Experiment—a small array of flexible, lightweight microwave transmitters aboard the Space Solar Power Demonstrator satellite that had launched in January. It was not a useful amount of power. Ali Hajimiri, the Caltech professor who co-directs the project, called it a detection, not a transmission. But it appeared at the expected time, at the expected frequency, with the correct frequency shift predicted by its travel distance from low Earth orbit. It was the first time anyone had demonstrated wireless energy transfer in space using flexible lightweight structures, and the first time detectable power had been beamed from orbit to Earth’s surface.
The concept is straightforward to state and staggeringly difficult to execute: put solar panels in space, where the sun shines 24 hours a day with no atmosphere, no weather, no seasons, and no night, then convert the collected energy to microwaves and beam it to receiver stations on the ground. Solar potential in orbit is roughly eight times greater per square meter than the best locations on Earth’s surface. The energy supply is effectively unlimited and perpetually available. The idea has existed since 1968, when Peter Glaser first described it. He was granted a U.S. patent for the method in 1973. Half a century later, what Caltech demonstrated on that Pasadena rooftop is still closer to a proof of concept than a power plant.
Why it’s hard
The engineering challenges cascade in every direction simultaneously, and each one is large enough to be a career-ending problem for an ordinary technology program.
Size is the first constraint. A commercially relevant space-based solar power array would need to be enormous—on the order of square kilometers. Caltech’s eventual design envisions individual units that fold into packages about one cubic meter and unfurl into flat squares roughly 50 meters per side, with solar cells on one face and microwave transmitters on the other. A full power station would be a constellation of these modular units. Some reference designs for operational systems run to 3.5 square miles of collector area in geostationary orbit, 36,000 kilometers above the equator, with similarly massive rectenna arrays on the ground to capture the incoming microwave beam. For context, the International Space Station has a total pressurized volume roughly equivalent to a five-bedroom house. The structures being discussed for space-based solar power are orders of magnitude larger.
Mass is the second. Getting material to orbit costs money, and the cost scales with weight. Caltech’s approach specifically targets this: their modular design uses ultralight, flexible structures that avoid the need for robotic in-space assembly of rigid components. The SSPD-1 prototype weighed 50 kilograms total. But scaling from a 50-kilogram demonstrator to a multi-gigawatt power station involves mass quantities of solar cells, structural materials, power electronics, and transmission hardware that must survive launch vibration, thermal cycling, radiation exposure, and micrometeorite impacts for decades of operation. China has announced plans to launch a 200-tonne space-based solar power station by 2035. Two hundred tonnes is a starting point, not a ceiling.
Transmission efficiency is the third. Microwaves spread as they travel—diffraction is physics, not engineering, and you can’t negotiate with it. The size of the beam spot on the ground is a function of the transmitter size and the frequency of the microwaves. Caltech’s MAPLE used a very small transmitter, which spread the power over a very large area, which is why the rooftop detection captured only a tiny fraction of the transmitted energy. Scaling up the transmitter reduces the spread but requires the enormous structures described above. Atmospheric absorption, weather interference, and conversion losses at both ends—electricity to microwaves in orbit, microwaves back to electricity on the ground—all take their cut. End-to-end efficiency from sunlight in orbit to electricity in the grid is a number that determines whether space-based solar is cheaper or more expensive than terrestrial alternatives, and current estimates remain unfavorable against rapidly falling ground-based solar costs.
Cost is the fourth and arguably the defining constraint. Recent deep-dive analyses commissioned by NASA and the European Space Agency have thrown cold water on near-term affordability. The ESA’s 2021 assessment suggested space-based solar power might become viable in the 2040s, but only with sustained investment and strong public-private cooperation. Launch costs have dropped dramatically with reusable rockets—SpaceX has fundamentally changed the economics of getting mass to orbit—but the total system cost for a multi-gigawatt power station, including manufacturing, launch, deployment, operation, maintenance, and ground infrastructure, remains formidable. The question isn’t whether the physics works. The question is whether the economics work before terrestrial solar, wind, and battery storage make the whole concept unnecessary.
What actually happened with Caltech’s mission
The SSPD-1 mission, which ran through most of 2023, tested three technologies. DOLCE demonstrated the deployment of a 1.8-meter-by-1.8-meter ultralight structure—the packaging and unfolding mechanism for future modular spacecraft. It deployed successfully despite two anomalies that gave the team, in Sergio Pellegrino’s words, “many new insights” into the structural challenges of ultralight deployable systems. ALBA tested 32 different types of photovoltaic cells over 240 days in orbit, including three entirely new classes of ultralight solar cells custom-fabricated at Caltech that had never been tested in space. And MAPLE demonstrated the wireless power transmission that made headlines.
The mission ended in November 2023 when the testbed stopped communications with Earth. The team’s assessment was measured: MAPLE proved the basic concept works—flexible lightweight structures can survive launch, deploy in orbit, and transmit a detectable beam to a ground station. It also revealed weaknesses. Hajimiri’s team pushed MAPLE to its limits deliberately, exposing failure modes that will shape the next generation of hardware. The mission accelerated what would normally be years of in-space testing into months, giving Caltech a feedback cycle that is unusually fast for space technology development.
Who else is working on it
Caltech isn’t alone. Japan’s JAXA demonstrated wireless microwave power transmission of 1.8 kilowatts over 50 meters on the ground in 2015. Mitsubishi Heavy Industries transmitted 10 kilowatts over 500 meters the same year. China has been building a testing base in Chongqing’s Bishan District and has announced progressively ambitious plans for space-based solar power stations, with CAST vice-president Li Ming stating China expects to be the first nation to build a working station with practical value. Researchers at King’s College London estimated in 2025 that space-based solar could provide Europe the majority of its renewable energy needs by 2050.
Aetherflux, a venture-funded startup that raised $50 million, proposed a constellation of small low Earth orbit satellites using infrared lasers instead of microwaves—a different transmission approach that allows smaller ground stations (5 to 10 meters in diameter versus the square-kilometer rectennas required for microwave systems). The company received partial support from the U.S. Department of Defense’s Operational Energy Capability Improvement Fund. Then, in December 2025, Aetherflux pivoted to space-based data centers, which tells you something about the current commercial viability of orbital power beaming.
The military angle deserves honest acknowledgment. Space-based solar power has dual-use implications that have shaped the field since the 1980s. The same technology that beams clean energy to a rectenna can, in principle, deliver directed energy to other targets. The 1980s Strategic Defense Initiative drew directly from early space solar power concepts. The 2025 Golden Dome missile defense program continues to fund orbital power generation and directed-energy transmission. The technology doesn’t just solve a clean energy problem. It creates a weapons capability. Both of these facts are part of the engineering reality.
Where it stands
Space-based solar power is the kind of technology that is simultaneously inevitable and permanently premature. The physics is unambiguous: space has more solar energy than Earth’s surface, and microwaves can transmit it through the atmosphere. The engineering is real: Caltech beamed detectable power from orbit to Pasadena. The economics are punishing: every analysis that accounts for full system costs, launch logistics, maintenance, and the pace of terrestrial renewable development concludes that the timelines are measured in decades, not years.
The honest framing is that space-based solar power is a hedge—a technology worth developing because there are plausible futures in which terrestrial renewables plus storage aren’t sufficient to meet global energy demand, and because the enabling technologies (lightweight structures, wireless power transmission, low-cost launch) have applications beyond solar power even if the full system never reaches commercial viability. It is not the next solar panel. It’s the generation after that, waiting for the cost curves to cross.
Caltech detected a signal on a rooftop in Pasadena. That’s a first. The distance between a detected signal and a powered civilization is approximately the same distance as the distance between the first telegraph and the internet—not a difference in kind, but a difference in scale that might take longer than the optimists think and less time than the skeptics assume.
We cover space-based solar power alongside fusion energy, quantum computing, and the full landscape of civilization-scale moonshot technologies across our Moonshot 2169 course—including why the best solar panel location on Earth isn’t on Earth.
