An internal AI tool served as a “springboard” for attackers to seize wallet developers’ computers, Binance says, during an operation spanning months that stole nearly $6.3 billion from the crypto platform. The company’s explanation still leaves a central security question open: how, exactly, did that tool let the attackers move from one employee’s system to another?
Binance’s report on the investigation, released Tuesday traces phishing, bogus job interviews and tampered software across three wallet-team employees. AI tools figure in two stages of the sequence, although the report does not show that AI itself caused the theft.
In June, a wallet developer clicked a malicious advertisement while attempting to install a popular AI coding tool. The download was a modified copy of Homebrew, a tool widely used by software engineers. According to Binance, it spotted the anomaly, gave the employee a replacement device and changed the associated credentials.
That response left an opening. Through the first attack, the hackers learned what the employee and colleagues on the wallet team did, information that helped them select another engineer to target.
A hacker claiming to be a project director approached this second engineer on a legitimate job-hunting platform. The proposed interview process began with an “HR representative” and continued with the “CEO” of a company that did not exist. The fraudsters sought to build trust by sharing a social media account featuring personal photos and videos.
The approach, Binance said, “was more advanced and deceptive than previously seen social engineering approaches.” In its coverage of the theft, Futurism raises the possibility that AI generated those photos and videos, while acknowledging that Binance’s report leaves this unclear. Any AI role at this stage is therefore speculative rather than confirmed.
For the next interview, the conversation moved to text. The engineer received instructions to complete and download an “assignment” with malicious code inside; that code executed immediately when installation started. Among its changes, the malware swapped a genuine development tool for a malicious copy capable of exchanging messages with the hackers and running commands. These events occurred in September.
Beginning in October, the attacker reached an internal AI tool through the compromised computer. Binance describes that tool as the “springboard” used to gain control over both the second engineer’s machine and the computer belonging to a third wallet developer. How the attackers could make that move is not explained. The report names the tool as part of the sequence but does not identify the security failure that made its use possible.
Next, the hackers signed into the third employee’s computer with credentials taken from the first wallet-team employee. There, they activated the malicious script that set the theft in motion. Why those credentials still played a role is another unanswered question, since Binance says it changed relevant credentials when it discovered the June incident.
From five wallets, the attackers moved 233.46 million XRP, a cryptocurrency, to wallets they controlled. They subsequently stole additional cryptocurrencies, including bitcoin. This could have been the most lucrative heist of all time, but that status has not been established.
Who carried out the attack is still uncertain. The FBI promptly attributed it to a notorious North Korean hacking outfit; Pyongyang denied any involvement. Binance reported that, before this breach, it had stopped “57 suspected North Korea-linked cyber threat cases”.
Futurism presents the theft as a warning about companies adopting AI coding tools. The narrower concern emerging from Binance’s explanation is how tampered developer software and stolen credentials worked alongside an internal AI tool to let attackers reach another wallet engineer’s computer. What permissions did the tool hold, and how were they used to cross between systems? The account of the breach still does not supply those details.
