Two weeks after North Korean hackers stole about $6.3 billion from cryptocurrency exchange Binance, internal investigations are clarifying what went wrong. According to one cyber expert, the genuine novelty in the attack was the use of Binance's own artificial intelligence models to help carry it out.
"The new thing here is not that the attacker had AI. It is that the victim did. On 20 October, the attacker used one of Binance's own internal AI tools as part of reaching two more engineers' computers," Ori Segal, CEO of Cybore, told The Jerusalem Post on Thursday. His company develops a system designed to detect suspicious activity during Zoom calls.
Segal said that, apart from that step, the entire operation was executed by classic means: a poisoned package manager, a fake recruiter, a malicious file, credential reuse and an over-broad deployment permission. "None of it required a model," he said, noting that while security professionals have spent years asking what hackers can do with AI, few have asked what hackers can do with the AI tools companies themselves use. "This is the first large incident where that question appears in the public record at all," he added.
Segal also warned that companies have few options for preventing such attacks or training staff to spot fraudulent job interviews, because there is no reliable guide for surviving them. "Being asked to build something, run it locally and pull in external libraries is what a coding interview is. A fake employer can have a website, a careers page and a set of plausible LinkedIn profiles, and we have no public information about what this one had. Anyone offering a reliable tell is guessing," he told the Post.
Instead of detection, Segal argued, prevention should focus on where code is run. Companies should ease the pressure for quick results and ask whether a task must be performed on a particular device and whether it carries infection risk. "Unvetted third-party code should never run on your work computer, and not on the machine holding your own personal projects either," he said, adding that "a disposable virtual machine costs nothing and removes the entire attack, whether or not you noticed anything was wrong."
Segal pointed to one further vulnerability he said has been largely overlooked: the job interview lure itself. An employee who suspects something went wrong faces a personal incentive to stay silent, because reporting the incident means admitting they were interviewing elsewhere on a company laptop. He stressed he was not suggesting that happened at Binance, saying he has no information either way, but called the incentive "structural" and said it exists in every exposed company and works in the attacker's favor.
"The fix is cheap and almost nobody has done it. Say out loud, in advance, that an employee who was job hunting and ran something they should not have will not be punished for reporting it. Until a company has said that, it should expect disclosure to arrive late and incomplete," he concluded.
According to Binance's initial report on November 3, hackers used a bogus job interview and software hidden in an advertisement to compromise the computers of two software engineers, then used a third engineer's computer to drain $6.3 billion in crypto in what could be the largest financial theft ever recorded. The exchange said the multi-stage operation began in June and set off a chain of security breaches that ended in the theft.
Binance said that even as it rushed to lock down its systems and freeze withdrawals after detecting the attack late on a Thursday, the hackers continued removing crypto from the exchange for nearly four hours after the first theft. Reuters contributed to the original report.
