Imagine being a developer at one of the world’s largest crypto companies. You sit down on an ordinary afternoon, search for a piece of software you need, and click on a link that looks familiar. Maybe it is a Google ad, not even the first result, but it appears relevant. You download the installer, run it, and go back to work. Months later, billions of dollars are gone—and that one tiny decision helped make it possible. That, according to Binance, is how the largest cryptocurrency theft in history began. The postmortem the company later published does not read like a Hollywood hacker thriller. There were no walls of scrolling code, no dramatic battle against a rogue AI, no split-second chase across cyberspace. Instead, the story sounds almost mundane: a malicious ad, a fake job interview, a hidden backdoor, and an internal company tool turned against its owners. The attackers combined several well-known hacking techniques over the course of months, and while Binance has been careful not to blame any single individual, the chain of events shows how small human mistakes can cascade into catastrophic financial losses. It started in June, when one developer downloaded a tampered installer. The installer was apparently spread through a malicious Google ad, a trick that remains surprisingly effective because people trust search results and advertisements by default. Binance says its security systems detected that first infection and replaced the worker’s device. That might have ended the story if this were a normal criminal group. But the attackers weren’t trying to steal something that day. They were conducting reconnaissance, taking notes about how the company’s engineers worked, which tools they used, and where the most valuable secrets were likely to live. The June breach wasn’t the heist. It was the scouting trip.
The next chapter began in September, with another engineer, another fake identity, and a much more patient trick. Someone pretending to be a recruiter reached out to an engineer, presumably with an attractive job offer. The attacker then did something that makes the scam so difficult to spot: they committed to the role. There were phone calls, several rounds of interviews, and social media pages that looked real enough to pass a quick background check. The conversation was probably tailored to the engineer’s interests, because the attackers had already learned a great deal about the company’s culture from the June breach. Over time, the engineer developed a sense of trust. The fake recruiter was friendly, plausible, and interested. Eventually, the recruiter asked the engineer to complete an interview assignment. That is normal enough. Software developers often receive coding tasks during the hiring process. But this assignment was not a test. It was malware, disguised as a technical challenge. The engineer downloaded it, probably ran the code as instructed, and in doing so handed the attackers a backdoor. The malware deleted a legitimate development tool called pyenv and replaced it with a malicious copy that could execute commands on the machine and maintain communication with an attacker-controlled server. According to Binance, this malware was a 0-day, meaning it exploited a weakness that no security vendor had seen before. It used clever obfuscation to hide its true purpose, and it went out of its way to erase evidence of its presence. The endpoint security tools that had caught the June attack did not catch this one. It was a second, much more dangerous intrusion, and the company didn’t know about it until it was already part of the final heist.
Between September and October, the attackers did not need to hurry. They had a backdoor into the company’s infrastructure and the kind of patience that separates serious cybercriminals from opportunistic thieves. By October 20, according to Binance, the attackers had managed to use the company’s own internal AI tool to gain control over cryptocurrency wallets belonging to the first engineer compromised in June, as well as a third employee. This detail is easy to skim past, but it deserves a pause. The internal AI tool was built to help Binance employees do their jobs, perhaps by automating tasks or providing access to data. The attackers did not try to steal the AI tool itself; they simply used the privileges that had been granted to it, or maybe to the employees who used it, to move from the outside to the center of the operation. They now had their eyes on wallets that contained enormous sums of money. Then, on October 29, they logged into the third employee’s computer and launched the final phase. By that point, the attack was no longer a subtle infiltration; it was an armed takeover. The company’s own systems were being used against it, and the funds began pouring out of its hot wallet. The entire operation, from the first tampered installer in June to the actual theft in late October, took months. The hackers had not acted impulsively. They studied the environment, chose the right people, waited for the right moment, and then moved.
After the theft, the obvious question is: why didn’t Binance stop the transfers? In a normal bank, a suspicious transaction can be halted by employees or fraud systems. In a cryptocurrency exchange, things are more complicated. The funds were held in a hot wallet, meaning a digital wallet connected to the internet and used for day-to-day withdrawals and deposits. To keep the hot wallet safe, Binance used Multi-Party Computation, a sophisticated security technique often called MPC. The idea behind MPC is to eliminate the single point of failure. Instead of one private key controlling the wallet, the ability to sign transactions is split among several parties. Think of a bank safety deposit box that requires two different keys to open. If one key is stolen, the thief still cannot get the box. MPC is meant to work that way. But on the night of the theft, this security feature worked against Binance. The attackers had obtained what Binance describes as malicious EIP-7702 authorizations. In simpler terms, they had managed to give themselves valid permission slips to move money. Because the wallet operated with MPC, no single private key could be revoked to cancel those authorizations. In a system with one key, you might be able to freeze the wallet by disabling that key. In an MPC system, the validity of the authorizations was judged collectively, and the collective decision was fooled. Binance says that unauthorized transactions continued for nearly two hours after the first one appeared. For users, those two hours were terrifying; for the hackers, they were an opportunity to drain as much as possible. By the time the transfer process was finally halted, the scale of the theft had already rewritten the record books.
According to the FBI, the attack was the work of a North Korean hacking unit called TraderTraitor. That name has become familiar to crypto security teams; North Korean hackers have been linked to a series of major cryptocurrency thefts, which are believed to help fund state priorities despite international sanctions. Binance says it has already prevented fifty-seven suspected cyber threat cases tied to North Korea, which suggests the company has spent years trying to block exactly this kind of threat. But this time the attackers got through. After the theft, the company tried hard to reassure users. CEO Richard Teng released a statement emphasizing that no user is bearing the loss. He said that about $2.85 billion in stolen assets had been frozen or intercepted, and that at that moment 98% of impacted tokens had been replaced, with the final 2% being completed. He also said that withdrawals had broadly resumed. In other words, the exchange was promising to make users whole, rather than passing the disaster onto the people who trusted it with their money. That is not a small thing. In previous crypto failures, customers have often been left waiting for years, unsure whether they would ever see their funds again. Binance appears to have absorbed the blow. At the same time, the phrase “frozen or intercepted” should make us stop and ask what those numbers actually mean in recoverable funds. Freezing an asset isn’t the same as putting it back into someone’s wallet. Still, the message was clear: the company was going to protect its customers, and it wanted the world to know.
After reading the postmortem, one is left with a strange feeling. The largest crypto theft in history is at least partly explained, but not fully understood. The company goes into technical detail about the malware, the backdoor, and the MPC wallet, but it offers much less detail about the two central questions that security researchers want answered. First, why did a tactic as well known as a fake interview assignment still work? For years, experts have warned that developers are prime targets for this exact kind of attack. It is common knowledge in security circles that an interview task asking you to download and run code from a stranger is dangerous. Yet this attack happened anyway, because in the middle of a real hiring process, with a recruiter who has spent days building trust, the warning signs are easy to miss. Second, what exactly is the internal AI tool that the attackers used, why did it have the ability to control wallets, and why wasn’t it more strictly protected? The postmortem describes it almost in passing, as if it were a minor detail. But it is not minor. An AI system that can be turned into a weapon for moving money is a critical risk. If the company doesn’t fully understand how the attackers were able to use the tool, then the same attack could happen again. More broadly, this heist shows that cybersecurity is not only about code. It is about people. The best encryption in the world cannot help if an engineer can be talked into running a malicious file. No firewall can stop an attacker who has been given the keys by a fake recruiter. The human instinct to trust, to help, to be liked, is powerful, and hackers know it. The lesson may be grim: the largest crypto theft was not accomplished by a brilliant superweapon, but by a series of quiet, patient deceptions. The only silver lining is that if the problem is human, the solution can also be human: better training, more skepticism, and a workplace culture where it is okay to stop and ask, “Wait, why is this person asking me to run this code?” Until then, the story of this theft is a warning, not a relief.