Binance traces record $6.3B crypto theft to North Korean fake job interview

How the attack unfolded

The record $6.3 billion theft from cryptocurrency exchange Binance on October 29 was the end point of a months-long, multi-stage campaign, according to an incident report the exchange published on Tuesday and detailed by The Register. Binance said the operation combined social engineering through fake job interviews, malware downloads, compromised internal AI tooling, endpoint takeover and what it described as an advanced method for authorizing transactions. In the report's words, it "was not a single isolated breach or a simple phishing incident."

The FBI attributed the theft over the weekend to TraderTraitor, a North Korean operation linked to the Lazarus Group. As The Register notes, the crew is practiced at job-market scams: sometimes posing as IT workers to get hired inside Western firms, sometimes playing the employer to trick engineers into loading malware onto corporate devices.

Reconnaissance through a poisoned Homebrew install

The groundwork was laid in June, when a Binance engineer downloaded a malicious Homebrew installation in what the exchange characterized as a supply chain attack. Security staff detected the malware, replaced the device and rotated the affected credentials, but not before the attacker learned the identity of a wallet DevOps engineer — referred to as Engineer A — and his colleagues on the wallet infrastructure team.

The fake interview

In September, a second software engineer was contacted through what Binance called a legitimate job-hunting platform by someone posing as a project director, who arranged a series of interviews with a fictitious company. During the first round, a live video call with a fake HR representative, the interviewer shared a social media account complete with personal photos and videos to build rapport — a touch Binance judged more deceptive than the social engineering it had previously encountered.

The second interview, with a person claiming to be the phony company's CEO, ran on a popular chat app. Near the end of the call, the engineer was told to download an assignment: a code repository laced with malware that executed on the work laptop during setup.

Malware inside the toolchain

According to the incident report, the payload replaced a legitimate development tool component — the pyenv version manager installed under Homebrew — with a malicious build that executed commands and opened a channel to an attacker-controlled server. The intruders also concealed the script well enough that Binance's endpoint security tools missed it initially.

On October 20, using the compromised machine together with an internal AI tool, the attackers reached the computers of Engineer A and another wallet developer, Engineer C. Engineer A's computer held a container deployment permission token, which the attackers discovered. On October 29 they entered Engineer C's machine, used Engineer A's stolen credentials to deploy a container, and triggered the script that launched the theft.

Remediation

Binance says it has rebuilt the compromised wallet system and the internal AI tool, replaced the affected devices, and decommissioned every credential and token assigned to the compromised engineers. It is working with law enforcement, blockchain intelligence firms and security researchers to trace and freeze the assets, has hired independent cybersecurity experts to review the incident and recommend safeguards, and is setting up a security research fund focused on AI security in blockchain settings.

Why it matters

This is being described as the largest cryptocurrency theft on record, and the attack chain explains why conventional defenses did not stop it. The initial compromises arrived through the ordinary furniture of a developer's working life — a package manager, a take-home coding assignment — and the final escalation relied on valid credentials rather than an exotic exploit. The role of the internal AI tool as a bridge to the wallet infrastructure team is a second warning: once a single endpoint fell, that tooling widened the blast radius. The practical takeaways for engineering organizations are to treat interview assignments and third-party installs as untrusted input, watch for tampering with installed toolchains, and not assume endpoint detection alone will catch a patient, well-resourced adversary.