Security ·
Binance explains $6.3 billion theft: fake Homebrew, fake job interview
On 3 November Binance published a preliminary report on the theft of about 6.34 billion dollars. It began with a tampered Homebrew installer that an engineer downloaded while looking for an AI coding tool, and continued with a fake job interview.
On 3 November Binance published a preliminary incident report on an attack that drained about 6.34 billion dollars from its wallets, according to the report. Outlets quote slightly different totals (from 6.3 to 6.4 billion), so we keep the figure from the exchange's own report. The theft took place on 29 October, but Binance says preparation reached back to June.
The first step was an engineer on the Wallet DevOps team. In June the engineer searched online for a popular AI coding tool and downloaded a tampered Homebrew installer, the package manager for Mac. This was not a compromise of the Homebrew project itself but a fake installer reached through search. Binance did not name the AI tool or the search engine involved.
The second step came in September. Another employee, taking part in a fake hiring process, downloaded a coding assignment onto a work computer. According to the report, the attackers then used the compromised machine and an internal AI tool to take over two more wallet infrastructure engineers by 20 October, obtaining credentials and a container deployment token. The FBI has attributed the theft to North Korea, naming the group it calls TraderTraitor, and the blockchain analysis firms TRM Labs and Chainalysis point the same way.
Binance says it has rebuilt its wallet systems and the internal AI tool, replaced the affected devices and revoked access. The report is preliminary and details may change. For companies outside crypto the pattern matters: the attack did not start in production infrastructure but on a developer laptop, and it abused trust in tools that people download themselves.
What it means for companies
A European company does not need to hold billions to face the same scenario: one developer who downloads a fake tool from a search result, or one employee who runs foreign code during a hiring task, is enough. What to do now: install developer tools only from official sites and repositories, never open candidate assignments on a work machine, separate developer laptops from production access and shorten token lifetimes. If you run internal AI tools or agents, give them minimal permissions and require human approval for critical actions, which is how we build custom AI agents. Our cyber security service covers access audits, staff training and monitoring.
Sources: BleepingComputer, The Star (Reuters)