Binance says $6.3 billion theft began with a fake job interview and a poisoned coding test

The exchange's timeline has four stages. In June, an engineer on the wallet team, setting up a coding tool, downloaded a tampered installer from an online advertisement instead of the company's usual channel. Binance caught the intrusion, replaced the computer and changed the credentials, but the intruders had already learned who worked on its wallet systems. Reports differ on the exact tool involved, one calling it an AI coding tool and another a popular developer package manager, so the safest description is a tampered installer for a developer tool.

In September, a different engineer, whom Binance calls Engineer B, was approached through a legitimate job-hunting website by a person posing as a project director. The engineer's public profile mentioned his Binance experience. A first-round interview was held with someone playing a human-resources representative, and a second with a person claiming to be the chief executive of a fictitious company. The engineer was then asked to download a coding assignment. The code ran during setup and evaded Binance's security software, which gave the attackers control of his computer.

On 20 October, the attackers used that computer and an internal Binance AI tool to take over the machines of two other wallet engineers, a wallet DevOps engineer and a wallet software engineer. The DevOps engineer's computer held a token that gave permission to deploy containers. On 29 October, the attackers used one engineer's credentials on the other's machine to reach the service that approves transfers. That let them move funds outside the exchange's normal withdrawal process.

Binance said AI played a limited role: the attack was mainly conventional phishing, and the internal AI tool was used only after the first laptop had been compromised, to gain access to other accounts. The finding matters because the tool turned one compromised laptop into three.

Recovery and the people question

Between roughly $2.5 billion and $2.85 billion of the stolen assets were frozen or intercepted, depending on the account, after stablecoin issuers Tether and Circle froze stolen tokens. Chainalysis chief executive Jonathan Levin has warned that the volume stolen exceeds what trusted laundering venues can handle, so the funds are likely to move through bridges, instant exchanges and swap services, which move value across blockchains or between coins and can make tracing harder. Those are the places where the frozen share could still grow.

Binance said that employees in roles that are not public-facing may not keep profiles on professional networking sites. It declined to say whether Engineer B had broken that policy or would be dismissed. The company said it had flagged earlier targeting of its staff through such sites and that in 2026 it had detected and prevented 57 hacking attempts linked to North Korea. This is the human side of the story: an engineer looking for a better job, a sophisticated adversary, and a company policy that mostly existed on paper.

The wider context supports the FBI's attribution as plausible, though Binance's own account does not independently prove it. North Korean operatives have long posed as job seekers and as recruiters to steal money for the regime's weapons programmes. The American intelligence community assessed earlier this year that Pyongyang nets at least $1 billion a year from cryptocurrency theft and other financial crime. The FBI also disrupted a separate North Korean laptop-farm operation this autumn, in which operatives used American-based laptops to appear to be working domestically.

What the case teaches

The most useful lesson is that the weakest point was not cryptography. It was a handful of engineers, their laptops and the permissions those machines held. Four practical points follow, and they apply well beyond exchanges.

First, treat any coding assignment from a recruiter or a prospective employer as untrusted code. Run it only in a disposable virtual machine with no access to work credentials, never on a work laptop, and verify that the company and the interviewer exist through a channel you find independently. Second, restrict what can be downloaded and executed on corporate machines. A security researcher quoted on the case said organisations need controls over what can be installed and over the permissions granted to users, and that applies equally to installers found through online ads. Third, limit the credentials that sit on a developer's computer. A deployment token left on a laptop is a master key for anyone who gets inside. Fourth, watch how internal AI tools are connected, because a tool that can reach many accounts is a convenient path for an intruder who has taken one.

For customers, the practical assurance is that reserves covered the loss. For the industry, the bigger message is that an exchange with large security budgets lost billions through two employees and a job offer. Until approvals for moving funds require independent hardware-backed sign-off from people who cannot be impersonated on a single compromised laptop, the same route will remain open to anyone with patience and a convincing fake company.