Radically different
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.
In February 2019, a group of scientists proposed a high-risk, cutting-edge, irresistibly exciting idea that the National Science Foundation should fund: making “mirror” bacteria.
These lab-created microbes would be organized like ordinary bacteria, but their proteins and sugars would be mirror images of those found in nature. Researchers believed they could reveal new insights into building cells, designing drugs, and even the origins of life.
But now, many of them have reversed course. They’ve become convinced that mirror organisms could trigger a catastrophic event threatening every form of life on Earth. Find out why they’re ringing alarm bells.
—Stephen Ornes
This story is from the next issue of our print magazine, which is all about nature. Subscribe now to read it when it lands this Wednesday.
Earlier this month, a GitHub project called Colleague Skill struck a nerve by claiming to “distill” a worker’s skills and personality—and replicate them with an AI agent. Though the project was a spoof, it prompted a wave of soul-searching among otherwise enthusiastic early adopters.
A number of tech workers told MIT Technology Review that their bosses are already encouraging them to document their workflows for automation via tools like OpenClaw. Many now fear that they are being flattened into code and losing their professional identity.
In response, some are fighting back with tools designed to sabotage the automation process.
—Caiwei Chen
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 The White House and Anthropic are working toward a compromise
The Trump administration says they had a “productive meeting.” (Reuters $)
+ Trump had ordered US agencies to phase out Anthropic’s tech. (Guardian)
+ Despite the blacklist, the NSA is using Anthropic’s new Mythos model. (Axios)
2 Palantir has unveiled a manifesto calling for universal national service
While denouncing inclusivity and “regressive” cultures. (TechCrunch)
+ It’s a summary of CEO Alex Karp’s book “The Technological Republic.” (Engadget)
+ One critic called the book “a piece of corporate sales material.“ (Bloomberg $)
3 Germany’s chancellor and largest company want looser AI rules
Chancellor Merz said industrial AI needs more regulatory freedom. (Reuters $)
+ Siemens says it plans to shift investments to the US if EU rules don’t change. (Bloomberg $)
+ Fractures over AI regulation are also emerging in the US. (MIT Technology Review)
4 Nvidia’s once-tight bond with gamers is cracking over AI
Consumer graphics cards are no longer the priority. (CNBC)
+ But generative AI could reinvent what it means to play. (MIT Technology Review)
5 Insurers are trying to exclude AI-related harms from their coverage
And escape legal liability for AI’s mistakes. (FT $)
+ AI images are being used in insurance scams. (BBC)
6 AI is about to make the global e-waste crisis much worse
And most of the trash will end up in non-Western countries. (Rest of World)
+ Here’s what we can do about it. (MIT Technology Review)
7 Tinder and Zoom have partnered with Sam Altman’s eye-scanning firm
To offer a “proof of humanity” badge to users. (BBC)
8 Islamist insurgents in West Africa are driving surging demand for drones
A Nigerian UAV startup is opening its first factory abroad in Ghana. (Bloomberg $)
9 Hundreds of fake pro-Trump AI influencers are flooding social media
In an apparent bid to hook conservative voters. (NYT)
10 A Chinese humanoid has smashed the human half-marathon record
Despite crashing into a railing near the end of the race. (NBC News)
+ Chinese tech firm Honor swept the podium spots. (Engadget)
+ Last year, humans won the race by a mile. (CNN)
Quote of the day
—Ben Cumming, head of communications at the AI safety nonprofit Future of Life Institute, tells the Washington Post that diverse public figures are endorsing a declaration of AI policy priorities.
One More Thing
The International Space Station will be decommissioned as soon as 2030, but the story of America in low Earth orbit (LEO) will continue.
Using lessons from the ISS, NASA has partnered with private companies to develop new commercial space stations for research, manufacturing, and tourism. If they are successful, these businesses will bring about a new era of space exploration: private rockets flying to private destinations.
They will also demonstrate a new model in which NASA builds infrastructure and the private sector takes it from there—freeing the agency to explore deeper and deeper into space. Read the full story.
—David W. Brown
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.)
+ Bask in thisadorable test of a dog’s devotion.
+ This vocal pitch trainer improves your singing straight from your browser.
+ Master international etiquette with this interactive guide to the world’s cultures.
+ Explore the networks of public figures with this intriguing interactive graph.
Roboticists used to dream big but build small. They’d hope to match or exceed the extraordinary complexity of the human body, and then they’d spend their career refining robotic arms for auto plants. Aim for C-3P0; end up with the Roomba.
The real ambition for many of these researchers was the robot of science fiction—one that could move through the world, adapt to different environments, and interact safely and helpfully with people. For the socially minded, such a machine could help those with mobility issues, ease loneliness, or do work too dangerous for humans. For the more financially inclined, it would mean a bottomless source of wage-free labor. Either way, a long history of failure left most of Silicon Valley hesitant to bet on helpful robots.
That has changed. The machines are yet unbuilt, but the money is flowing: Companies and investors put $6.1 billion into humanoid robots in 2025 alone, four times what was invested in 2024.
What happened? A revolution in how machines have learned to interact with the world.
Imagine you’d like a pair of robot arms installed in your home purely to do one thing: fold clothes. How would it learn to do that? You could start by writing rules. Check the fabric to figure out how much deformation it can tolerate before tearing. Identify a shirt’s collar. Move the gripper to the left sleeve, lift it, and fold it inward by exactly this distance. Repeat for the right sleeve. If the shirt is rotated, turn the plan accordingly. If the sleeve is twisted, correct it. Very quickly the number of rules explodes, but a complete accounting of them could produce reliable results. This was the original craft of robotics: anticipating every possibility and encoding it in advance.
Around 2015, the cutting edge started to do things differently: Build a digital simulation of the robotic arms and the clothes, and give the program a reward signal every time it folds successfully and a ding every time it fails. This way, it gets better by trying all sorts of techniques through trial and error, with millions of iterations—the same way AI got good at playing games.
The arrival of ChatGPT in 2022 catalyzed the current boom. Trained on vast amounts of text, large language models work not through trial and error but by learning to predict what word should come next in a sentence. Similar models adapted to robotics were soon able to absorb pictures, sensor readings, and the position of a robot’s joints and predict the next action the machine should take, issuing dozens of motor commands every second.
This conceptual shift—to reliance on AI models that ingest large amounts of data—seems to work whether that helpful robot is supposed to talk to people, move through an environment, or even do complicated tasks. And it was paired with other ideas about how to accomplish this new way of learning, like deploying robots even if they aren’t yet perfect so they can learn from the environment they’re meant to work in. Today, Silicon Valley roboticists are dreaming big again. Here’s how that happened.
A movable social robot carried out conversations long before the age of LLMs.
An MIT robotics researcher named Cynthia Breazeal introduced an armless, legless, faceless robot called Jibo to the world in 2014. It looked, in fact, like a lamp. Breazeal’s aim was to create a social robot for families, and the idea pulled in $3.7 million in a crowdsourced funding campaign. Early preorders cost $749.
The early Jibo could introduce itself and dance to entertain kids, but that was about it. The vision was always for it to become a sort of embodied assistant that could handle everything from scheduling and emails to telling stories. It earned a number of devoted users, but ultimately the company shut down in 2019.
In retrospect, one thing that Jibo really needed was better language capabilities. It was competing against Apple’s Siri and Amazon’s Alexa, and all those technologies at the time relied on heavy scripting. In broad terms, when you spoke to them, software would translate your speech into text, analyze what you wanted, and create a response pulled from preapproved snippets. Those snippets could be charming, but they were also repetitive and simply boring—downright robotic. That was especially a challenge for a robot that was supposed to be social and family oriented.
What has happened since, of course, is a revolution in how machines can generate language. Voice mode from any leading AI provider is now engaging and impressive, and multiple hardware startups are trying (and failing) to build products that take advantage of it.
But that comes with a new risk: While scripted conversations can’t really go off the rails, ones generated by AI certainly can. Some popular AI toys have, for example, talked to kids about how to find matches and knives.
A robot hand trained with simulations tries to model the unpredictability and variation of the real world.
By 2018, every leading robotics lab was trying to scrap the old scripted rules and train robots through trial and error. OpenAI tried to train its robotic hand, Dactyl, virtually—with digital models of the hand and of the palm-size cubes Dactyl was supposed to manipulate. The cubes had letters and numbers on their faces; the model might set a task like “Rotate the cube so the red side with the letter O faces upward.”
Here’s the problem: A robotic hand might get really good at doing this in its simulated world, but when you take that program and ask it to work on a real version in the real world, the slight differences between the two can cause things to go awry. Colors might be slightly different, or the deformable rubber in the robot’s fingertips could turn out to be stretchier than it was in simulation.
The solution is called domain randomization. You essentially create millions of simulated worlds that all vary slightly and randomly from one another. In each one the friction might be less, or the lighting more harsh, or the colors darkened. Exposure to enough of this variation means the robots will be better able to manipulate the cube in the real world. The approach worked on Dactyl, and one year later it was able to use the same core techniques to do something harder: solving Rubik’s Cubes (though it worked only 60% of the time, and just 20% when the scrambles were particularly hard).
Still, the limits of simulation mean that this technique plays a far smaller role today than it did in 2018. OpenAI shuttered its robotics effort in 2021 but has recently started the division up again—reportedly focusing on humanoids.
Training on images from across the internet helps robots translate language into action.
Around 2022, Google’s robotics team was up to some strange things. It spent 17 months handing people robot controllers and filming them doing everything from picking up bags of chips to opening jars. The team ended up cataloguing 700 different tasks.
The point was to build and test one of the first large-scale foundation models for robotics. As with large language models, the idea was to input lots of text, tokenize it into a format an algorithm could work with, and then generate an output. Google’s RT-1 received input about what the robot was looking at and how the many parts of the robotic arm were positioned; then it took an instruction and translated it into motor commands to move the robot. When it had seen tasks before, it carried out 97% of them successfully; it succeeded at 76% of the instructions it hadn’t seen before.
The second iteration, RT-2, came out the following year and went even further. Instead of training on data specific to robotics, it went broad: It trained on more general images from across the internet, like the vision-language models lots of researchers were working on at the time. That allowed the robot to interpret where certain objects were in the scene.
“All these other things were unlocked,” says Kanishka Rao, a roboticist at Google DeepMind who led work on both iterations. “We could do things now like ‘Put the Coke can near the picture of Taylor Swift.’”
In 2025, Google DeepMind further fused the worlds of large language models and robotics, releasing a Gemini Robotics model with improved ability to understand commands in natural language.
An AI model that allows robotic arms to act like coworkers.
In 2017, before OpenAI shuttered its first robotics team, a group of its engineers spun out a project called Covariant, aiming to build not sci-fi humanoids but the most pragmatic of all robots: an arm that could pick up and move things in warehouses. After building a system based on foundation models similar to Google’s, Covariant deployed this platform in warehouses like those operated by Crate & Barrel and treated it as a data collection pipeline.
By 2024, Covariant had released a robotics model, RFM-1, that you could interact with like a coworker. If you showed an arm many sleeves of tennis balls, for example, you could then instruct it to move each sleeve to a separate area. And the robot could respond—perhaps predicting that it wouldn’t be able to get a good grip on the item and then asking for advice on which particular suction cups it should use.
This sort of thing had been done in experiments, but Covariant was launching it at significant scale. The company now had cameras and data collection machines in every customer location, feeding back even more data for the model to train on.
It wasn’t perfect. In a demo in March 2024 with an array of kitchen items, the robot struggled when it was asked to “return the banana” to its original location. It picked up a sponge, then an apple, then a host of other items before it finally accomplished the task.
It “doesn’t understand the new concept” of retracing its steps, cofounder Peter Chen told me at the time. “But it’s a good example—it might not work well yet in the places where you don’t have good training data.”
Chen and fellow founder Pieter Abbeel were soon hired by Amazon, which is currently licensing Covariant’s robotics model (Amazon did not respond to questions about how it’s being used, but the company runs an estimated 1,300 warehouses in the US alone).
Companies are putting this humanoid to the test in real-world settings.
The new investment dollars flowing to robotics startups are aimed largely at robots shaped not like lamps or arms but like people. Humanoid robots are supposed to be able to seamlessly enter the spaces and jobs where humans currently work, avoiding the need to retool assembly lines to accommodate new shapes such as giant arms.
It’s easier said than done. In the rare cases where humanoids appear in real warehouses, they’re often confined to test zones and pilot programs.
That said, Agility’s humanoid Digit appears to be doing some real work. The design—with exposed joints and a distinctly unhuman head—is driven more by function than by sci-fi aesthetics. Amazon, Toyota, and GXO (a logistics giant with customers like Apple and Nike) have all deployed it—making it one of the first examples of a humanoid robot that companies see as providing actual cost savings rather than novelty. Their Digits spend their days picking up, moving, and stacking shipping totes.
The current Digit is still a long way from the humanlike helper Silicon Valley is betting on, though. It can lift only 35 pounds, for example—and every time Agility makes Digit stronger, its battery gets heavier and it has to recharge more often. And standards organizations say humanoids need stricter safety rules than most industrial robots, because they’re designed to be mobile and spend time in proximity to people.
But Digit shows that this revolution in robot training isn’t converging on a single method. Agility relies on simulation techniques like those OpenAI used to train its hand, and the company has worked with Google’s Gemini models to help its robots adapt to new environments. That’s where more than a decade of experiments have gotten the industry: Now it’s building big.
There’s a fault line running through enterprise AI, and it’s not the one getting the most attention. The public conversation still tracks foundation models and benchmarks—GPT versus Gemini, reasoning scores, and marginal capability gains. But in practice, the more durable advantage is structural: who owns the operating layer where intelligence is applied, governed, and improved. One model treats AI as an on-demand utility; the other embeds it as an operating layer—the combination of operation software, data capture, feedback loops and governance that sits between models and real work—that compounds with use.
Model providers like OpenAI and Anthropic sell intelligence as a service: you have a problem, you call an API, you get an answer. That intelligence is general-purpose, largely stateless, and only loosely connected to the day-to-day operations where decisions are made. It’s highly capable and increasingly interchangeable. The distinction that matters is whether intelligence resets on every prompt or accumulates over time.
Incumbent organizations, by contrast, can treat AI as an operating layer: instrumentation across operations, feedback loops from human decisions, and governance that turns individual tasks into reusable policy. In that setup, every exception, correction, and approval becomes a chance to learn—and intelligence can improve as the platform absorbs more of the organization’s work. The organizations most likely to shape the enterprise AI era are those that can embed intelligence directly into operational platforms and instrument those platforms so work generates usable signals.
The prevailing narrative says nimble startups will out-innovate incumbents by building AI-native from scratch. If AI is primarily a model problem, that story holds. But in many enterprise domains, AI is a systems problem—integrations, permissions, evaluation, and change management—where advantage accrues to whomever already sits inside high-volume, high-stakes operations and converts that position into learning and automation.
Traditional services organizations are built on a simple architecture: humans use software to do expert work. Operators log into systems, navigate operations, make decisions, and process cases. Technology is the medium. Human judgment is the product.
An AI-native platform inverts this. It ingests a problem, applies accumulated domain knowledge, executes autonomously what it can with high confidence, and routes targeted sub-tasks to human experts when the situation demands judgment that the system can’t yet reliably provide.
But inverting human-AI interaction isn’t just a UI redesign—it requires raw material. It’s only possible when the platform is built on a foundation of domain expertise, behavioral data, and operational knowledge accumulated over years.
AI-native startups begin with a clean architectural slate and can move quickly. What they can’t easily manufacture is the raw material that makes domain AI defensible at scale:
Services companies already have all three. But these ingredients aren’t moats on their own. They become an advantage only when a company can systematically convert messy operations into AI-ready signals and institutional knowledge—then feed the results back into operations so the system keeps improving.
In most services organizations, expertise is tacit and perishable. The best operators know things they cannot easily articulate: heuristics developed over the years, edge-case intuitions, and pattern recognition that operate below the level of conscious reasoning.
At Ensemble, the strategy for addressing this challenge is knowledge distillation. The systematic conversion of expert judgment and operational decisions into machine-readable training signals.
In health-care revenue cycle management, for example, systems can be seeded with explicit domain knowledge and then deepen their coverage through structured daily interaction with operators. In Ensemble’s implementation, the system identifies gaps, formulates targeted questions, and cross-checks answers across multiple experts to capture both consensus and edge-case nuance. It then synthesizes these inputs into a living knowledge base that reflects the situational reasoning behind expert-level performance.
Once a system is constrained enough to be trusted, the next question is how it gets better without waiting for annual model upgrades. Every time a skilled operator makes a decision, they generate more than a completed task. They generate a potential labeled example—context paired with an expert action (and sometimes an outcome). At scale, across thousands of operators and millions of decisions, that stream can power supervised learning, evaluation, and targeted forms of reinforcement—teaching systems to behave more like experts in real conditions.
For example, if an organization processes 50,000 cases a week and captures just three high-quality decision points per case, that’s 150,000 labeled examples every week without creating a separate data-collection program.
A more advanced human-in-the-loop design places experts inside the decision process, so systems learn not just what the right answer was, but how ambiguity gets resolved. Practically, humans intervene at branch points—selecting from AI-generated options, correcting assumptions, and redirecting operations. Each intervention becomes a high-value training signal. When the platform detects an edge case or a deviation from the expected process, it can prompt for a brief, structured rationale, capturing decision factors without requiring lengthy free-form reasoning logs.
The goal is to permanently embed the accumulated expertise of thousands of domain experts—their knowledge, decisions, and reasoning—into an AI platform that amplifies what every operator can accomplish. Done well, this produces a quality of execution that neither humans nor AI achieve independently: higher consistency, improved throughput, and measurable operational gains. Operators can focus on more consequential work, supported by an AI that has already completed the analytical groundwork across thousands of analogous prior cases.
The broader implication for enterprise leaders is straightforward. Advantages in AI won’t be determined by access to general-purpose models alone. It will come from an organization’s ability to capture, refine, and compound what it knows, its data, decisions, and operational judgment, while building the controls required for high-stakes environments. As AI shifts from experimentation to infrastructure, the most durable edge may belong to the companies that understand the work well enough to instrument it and can turn that understanding into systems that improve with use.
This content was produced by Ensemble. It was not written by MIT Technology Review’s editorial staff.
From inside a money-laundering center in Cambodia, an employee opens a popular Vietnamese banking app on his phone. The app asks him to upload a photo associated with the account, so he clicks on a picture of a 30-something Asian man.
Next, the app requests to open the camera for a video “liveness” check. The scammer holds up a static image of a woman bearing no resemblance to the man who owns the account. After a 90-second wait—as the app tells him to readjust the face inside the frame—he’s in.
The exploit he’s demonstrating, in a video shared with me by a cyberscam researcher named Hieu Minh Ngo, is possible thanks to one of a growing range of illicit hacking services, readily available for purchase on Telegram, that are designed to break “Know Your Customer” (KYC) facial scans.
These banking and crypto safeguards are supposed to confirm that an account belongs to a real person, and that the user’s face matches the identity documents that were provided to open the account. But scammers are bypassing them in order to open mule accounts and launder money. Rather than using a live phone camera feed for a liveness check, the hacks typically deploy a tool known as a virtual camera. Users can replace the video stream with other videos or photos—depicting a real or deepfake person or even an object.
As financial institutions enact enhanced security measures aimed at stopping cyberscammers, these workarounds are the latest round in the cat-and-mouse game between criminal operators and the financial services industry.
Over the course of a two-month investigation earlier this year, MIT Technology Review identified 22 Chinese-, Vietnamese-, and English-language public Telegram channels and groups advertising bypass kits and stolen biometric data. The software kits use a variety of methods to compromise phone operating systems and banking applications, claiming to enable users to get around the compliance checks imposed by financial institutions ranging from major crypto exchanges such as Binance to name-brand banks like Spain’s BBVA.
“Specializing in bank services—handling dirty money,” reads the since-deleted Telegram bio of the program used by the Cambodian launderer, complete with a thumbs-up emoji. “Secure. Professional. High quality.” Some of the channels and groups had thousands of subscribers or members, and many posted bullet points listing their services (“All kinds of KYC verification services”; “It’s all smooth and seamless”) alongside videos purporting to show successful hacks.
Telegram says that after reviewing the accounts, it removed them for violating its terms of service. But such online marketplaces proliferate easily, and multiple channels and groups advertising similar tools remain active.
The rise in KYC bypasses has occurred alongside an expansion of a global industry in “pig-butchering” cyberscams. Crypto platforms and banks around the world are facing increasing scrutiny over the flow of illegally obtained money, including profits from such scams, through their platforms. This has prompted tightened banking regulations in countries such as Vietnam and Thailand, where governments have increased customer verification and fraud monitoring requirements and are pushing for stronger anti-money-laundering safeguards in the crypto industry.
Chainalysis, a US blockchain analysis firm, estimates that around $17 billion was stolen in 2025 in crypto scams and fraud, up from $13 billion in 2024. The United Nations Office on Drugs and Crime, meanwhile, warned in a recent report that the expansion of Asian scam syndicates in Africa and the Pacific has helped the industry “dramatically scale up profits.”
That combination of factors—more scrutiny, but also more revenue—has vaulted KYC bypasses to the center of the online marketplace for cyberscam and casino money launderers. Although estimates vary, cybersecurity researchers say these kinds of attacks are rising: The biometrics verification company iProov estimated that virtual-camera attacks were more than 25 times as common worldwide 2024 than in 2023, while Sumsub, a company providing KYC services, reported that “sophisticated” or multi-step fraud attempts, including virtual-camera bypasses, almost tripled last year among its clients.
Three financial institutions that were named as targets on such Telegram channels—the world’s largest crypto exchange, Binance, as well as BBVA and UK-based Revolut—told me they’re aware of such bypasses and emphasize that they’re an industry-wide challenge. A spokesperson from Binance said it has “observed attempts of this nature to circumvent our controls,” adding that “we have successfully prevented such attacks and remain confident in our systems.” BBVA and Revolut also declined to comment on whether their safeguards had been breached.
It’s difficult to estimate success rates, because companies may not be aware of bypasses—or report them—until later. “What’s important is what we don’t see,” Artem Popov, Sumsub’s head of fraud prevention products, told me, referring to attacks that go undetected. “There’s always part of the story where it might be completely hidden from our eyes, and from the eyes of any company in the industry, using any type of KYC provider.”
Advertisements for the exploits appear simple enough, but on the back end, building a successful bypass is complex and often involves multiple methods. Some channels offer to jailbreak a physical phone so that scammers can trigger the use of a virtual camera (VCam) instead of the built-in one whenever they’d like. Other hacks inject code known as a “hooking framework” into a financial institution’s app that triggers the VCam to open. Either way, VCams can be used to dupe KYC safeguards with images or videos that replace genuine, live video of the account’s owner.
Sergiy Yakymchuk, CEO of Talsec, a cybersecurity company that primarily serves financial institutions, reviewed details from the Telegram channels identified by MIT Technology Review and says they are consistent with successful tactics used against his banking and crypto clients. His team received help requests from banks and exchanges for roughly 30 VCam-based hacks over the past year, up from fewer than 10 in 2023.
Increasingly, hackers compromise both the phone itself and the code of the financial institutions’ apps before feeding the virtual camera a mix of stolen biometrics and deepfakes, Yakymchuk says.
“Some time ago, it was enough to decompile the app of a bank and distribute this on Telegram, and that was everything you needed,” he says. “Now it’s not enough, because you have KYC—and more and more things are needed.”
For money launderers, KYC bypasses have “become essential for everything right now—because scam compounds need to move money,” says Ngo, the researcher who shared the demo video. A convicted former hacker who became a cybersecurity advisor for the Vietnamese government, Ngo now runs an anti-scam nonprofit and helps law enforcement investigate money laundering.
He describes how the process works in the case of pig-butchering scams: Funds originating with victims are received into bank accounts controlled or rented by a money-laundering network, known colloquially as “water houses.” Money launderers use KYC bypasses to access the accounts and quickly redistribute the profits before converting them into digital assets—typically in the form of the stablecoin Tether, a type of cryptocurrency that is pegged to the US dollar.
These transactions often happen in seconds, under tightly orchestrated management. “They know, very clearly, the flow of how the banks verify or authenticate accounts,” Ngo says.
The growth of cyberscam money laundering has led to heightened scrutiny of financial institutions. In 2023, Binance pleaded guilty in US federal courts to operating without anti-money-laundering safeguards. Donald Trump pardoned former Binance CEO Chaopeng Zhao last October.
Recent analysis from the International Consortium of Investigative Journalists found that after Zhao’s guilty plea, more than $400 million continued to move to Binance from Huione Group, a Cambodia-based firm that the US sanctioned after the Treasury Department deemed it a “critical node” for money laundering in pig-butchering scams.
Binance says it has “state-of-the-art security systems” that prevented billions in fraud losses and that the company processed more than 71,000 law enforcement requests in 2025.
But John Griffin, a finance and blockchain expert at the University of Texas at Austin, does not think the exchanges are sufficiently secure. “Even though they have all this press about ‘Oh, yes, we’ve changed this and that’—well, the proof is in the pudding. The criminals are still using your exchange,” Griffin told me of the industry at large. “So there must be holes.” (Binance says it “objects to the dubious findings” of Griffin’s work tracking the flow of criminal profits across exchanges like Binance, Huobi, OKX, and Tokenlon, calling it “misleading at best and, at worst, wildly inaccurate.”)
Binance also pointed out that some purported bypass services are themselves scams, casting doubt on whether successful bypasses are as widespread as the Telegram marketplace may suggest. Engaging with such services “exposes individuals to significant security risks,” a spokesperson said. “Even where access appears to be granted, accounts are often already restricted by internal detection and compliance controls, rendering them nonfunctional for trading or withdrawals.”
Regulators around the world are trying to catch up. In Thailand, where citizens’ bank accounts regularly serve as money mules for cyberscams based in neighboring Myanmar and Cambodia, new legislation has enhanced KYC monitoring, limited daily transactions, and strengthened oversight bodies’ ability to suspend accounts. The US money-laundering regulator, the Financial Crimes Enforcement Network, issued a warning against KYC deepfakes and the use of VCams in late 2024, encouraging platforms to track broader transaction patterns to identify money laundering.
For scammers, any new security or reporting requirements will make bypasses harder, but “it’s not going to stop them,” Ngo says. “It’s just a matter of time.”