Skip to content
NewsIncident

Binance details how a hijacked internal AI tool and a fake job interview enabled its $6.34 billion theft

· by

The short version

Binance's November 3 incident update is the first detailed account of the attack chain behind its record crypto theft: a trojanized Homebrew install, a months-long fake recruiter ruse, and an internal AI tool used to pivot between compromised employees.

Binance details how a hijacked internal AI tool and a fake job interview enabled its $6.34 billion theft

Binance's record crypto theft has already been public for days. What changed on November 3 is that Binance finally explained how the attack worked, and the mechanism running through most of it is an internal AI tool the attackers never had to break into on their own, because they inherited access to it from an employee they had already compromised.

What Binance disclosed

Binance said the theft, which it estimates at $6.34 billion, began in June 2026 when a Wallet DevOps engineer searched for a popular AI coding tool and installed a trojanized Homebrew package, exposing details about his role and teammates to the attacker before Binance rotated his credentials, according to The Hacker News. In September, a second engineer was approached on a legitimate job site by someone posing as a company's Project Director. A first-round interview with a fake HR representative and a second-round interview with a fake CEO built enough trust that the engineer downloaded a bogus take-home assignment, and the malicious code inside it executed on setup, per Security Affairs' account. On October 20, the attacker used that compromised machine to abuse an internal AI tool and gain control of two more employees, including the original June target, who still had a container-deployment permission token sitting on his computer. On October 29 between 17:15 and 19:56, the attacker used those stolen credentials to deploy a container, run a malicious script, and reach Binance's internal wallet signing service, bypassing its normal transaction-authorization process.

Internal AI tools are now part of the employee-compromise attack surface

Binance's account does not describe a flaw in the AI model itself. The internal tool worked exactly as designed; it simply operated with whatever reach the already-compromised engineer's account carried, and the attacker used that reach to jump from one employee to the next. That distinction is the practical takeaway for any organization running an internal AI assistant with broad access to employee systems: the tool's own security does not matter if it inherits the permissions of a single phished account. Binance's response was to rebuild the tool and its access controls, not to patch a model, which signals where the company sees the actual gap.

Context and reactions

The FBI has linked the theft to the North Korea-affiliated TraderTraitor group, also tracked as Lazarus Group and APT38, per The National. Cybercrime researcher Geoff White noted on LinkedIn that fake-recruiter lures have a long history with this cluster: the 2016 Bangladesh Bank theft and the 2022 Axie Infinity breach both started the same way, cited in Security Affairs' writeup. Binance has said it will absorb the losses rather than pass them to customers and resumed withdrawals on November 1, and it told The Hacker News it has foiled 57 suspected North Korea-linked attempts against the company so far this year.

What to watch next

Binance has said a fuller independent technical postmortem, run with outside security reviewers, is still coming, and whether that review names the AI tool involved or discloses a different root cause than the November 3 account will matter more than the dollar figure already reported last week. Also worth tracking is whether any other exchange discloses a similar internal-AI-tool pivot, since Binance's own history of 57 foiled attempts this year suggests the fake-job-interview playbook is being run against the industry continuously, not just once.

Sources