Binance traces $6.3 billion breach to lateral movement via an internal AI tool

Binance disclosed that attackers who stole a record $6.3 billion from the exchange used fake job interviews and a compromised internal AI tool to move through its systems before draining its crypto wallets.

The exchange suspended withdrawals for customers worldwide after the theft, restoring service by the following Sunday.

What happened

Initial findings published by Binance said hackers gained access to its systems through a sophisticated, multi-stage campaign.

One engineer was contacted through a legitimate job-hunting platform and invited to bogus job interviews with fake company representatives.

The engineer was later instructed to download malware disguised as a coding assignment, which deployed malicious code using obfuscation techniques that evaded detection.

Binance said hackers then leveraged the engineer's compromised computer and an internal AI tool to gain control over two other engineers' laptops and execute the next steps of the attack.

Impact

Binance said the attackers used malicious software downloads to compromise the computers of its engineers and gain control over tools to drain its wallet of billions of dollars worth of crypto.

It is thought to be the largest crypto hack to date.

Response

Binance called the episode a serious and highly sophisticated security incident and apologised, saying this should not have happened.

It said it had set aside more than 100% of the funds needed to cover the loss, with 98% of the token assets impacted replenished as of Saturday.

US crime agencies, including the FBI and Secret Service, have been assisting Binance in tracing the crypto taken from the platform's own wallets.

Cyber analysts said the hack bore the signs of having been orchestrated by North Korea, a claim Pyongyang denied on Tuesday.

Lessons for security leaders

An internal tool that links engineers across teams can become a path for lateral movement the moment one identity on it is compromised, so such tools deserve monitoring as a trust boundary rather than as an ordinary application. Recruitment-themed social engineering aimed at engineers with privileged access warrants the same scrutiny given to external vendor onboarding, including unsolicited interview requests and take-home coding assignments. Rotating a single compromised credential is not enough; security teams should also map what access or information that identity could have reached through shared internal tooling before treating an incident as contained.