Binance Traces $6.34 Billion Breach to a Fake Job Interview and a Tampered AI Coding Tool Package

Binance says about $6.34 billion was taken after attackers reached engineers' laptops through fake job interviews and a tampered download of an AI coding tool. The exchange's November 3 update describes the loss as "affected" assets. It says it discovered the outflow on October 29 and that about $2.85 billion had been frozen or intercepted by November 3. If the gross figure holds, it is about four times the $1.5 billion taken from Bybit in February 2025.

According to BleepingComputer and Help Net Security, the attack had several stages. In June, a Wallet DevOps engineer installed a tampered Homebrew package while looking for a popular AI coding tool. In September, a person posing as a project director approached another engineer on a legitimate job site and had the engineer download a malicious "coding assignment" in a second-round interview. On October 20, the attackers used a compromised machine and an internal Binance AI tool to reach two more wallet engineers. Binance has not explained how the AI tool enabled that access.

The FBI reportedly attributed the attack to TraderTraitor, a North Korean state group. Binance has not tied the compromise to a named campaign, though the methods match documented North Korean tradecraft. The attribution comes from reported statements, not a public FBI filing.

New research targets one of these entry points. A November 5 paper by Hala Ali and Andrew Case, ModelShield, proposes runtime detection of malicious machine-learning models. It links observed system-level behavior to specific model operations during loading and inference, and it supports 17 frameworks and 36 formats. The authors argue that current scanners fail because they rely on static analysis and function denylists. An attacker can hide a payload from static analysis, but the payload still has to act when it runs.