Binance Hack Traced to Fake Job Interview and AI Tool Abuse

Binance has published its first detailed account of the hack that drained $6.34 billion on October 29. The report traces the intrusion back five months, to a tampered download in June and a job offer that never existed. Three engineers fell before the attackers reached the internal wallet signing service. An internal AI tool carried them between machines, and the FBI hastied the Binance hack to North Korean unit TraderTraitor.

How Binance Traced the Hack Back to June

The groundwork started in June, when an engineer on the Wallet DevOps team searched for a popular AI coding tool. He found it through a search engine and installed it using a tampered Homebrew package. The installer swapped a legitimate development component for a modified version that could run commands on his machine. Binance caught the anomaly, pulled the device and rotated the related credentials.

That response looked sufficient at the time, but it did not undo the damage. By then the attackers had already mapped his role and his colleagues on the wallet infrastructure team. The cleanup closed the access route and left the reconnaissance intact. That target list is what made the rest of the Binance hack possible.

Engineers install packages constantly, and poisoned search results sit alongside legitimate ones. One tampered installer reaches the people with the most direct access to production systems. Binance closed the June incident, but the hack that followed grew directly out of it.

September: Two Interviews and a Coding Assignment

Three months later the attackers came back, and they arrived through the hiring process. A second software engineer received a message on a legitimate job-hunting platform from someone posing as a Project Director. The company did not exist, but the pitch held up well enough to reach a first-round interview.

A supposed HR representative ran that interview as a live video call. She shared a social media account stocked with personal photos and videos, which built enough trust to move things forward. A second round followed over a chat app, this time with a person claiming to be the chief executive. It ended with a request any developer would recognise: download and complete a coding assignment.

The assignment contained a malicious code repository, and the malware inside it used a zero-day that endpoint security did not flag. The engineer’s work computer fell quietly, with no alert raised and no reason for anyone to look. Binance has detected and blocked 57 suspected North Korea-linked incidents across 2026, including other fake interviews and compromised packages. The hack that worked was the one Binance did not catch in time.

Why Recruitment Lures Keep Working

Fake recruitment campaigns work because they catch people when their guard is already down. A job applicant expects files from strangers, video calls with people they have never met, and technical exercises on short notice. Every step that would look suspicious elsewhere reads as routine inside a hiring process.

The attackers built this lure in layers and took their time. Two interviews, a real video call, a populated social profile and a plausible job title all came before the payload. They invested weeks of human effort before asking for anything, and that patience separates state-sponsored work from commodity phishing.

October: An Internal AI Tool Becomes the Pivot

On October 20, the attackers used the compromised machine and an internal Binance AI tool to reach two more systems. They took the original DevOps engineer’s device and that of a third wallet engineer. Those systems handed over deployment credentials for the wallet infrastructure. Binance has not explained how the attackers abused the AI tool, which leaves the biggest gap in its hack report.

This detail carries weight well beyond one exchange and one incident. Internal AI tooling holds broad access to code, infrastructure and employee systems, yet rarely gets the scrutiny applied to production services. An attacker who controls one workstation and can reach that tooling moves across the environment while looking like ordinary use.

Binance has rebuilt the internal AI tool along with its access controls. That choice suggests the design itself contributed to the Binance hack rather than one misconfiguration. Security teams rolling out similar tooling should treat it as production infrastructure from the start.

October 29: From Stolen Credentials to the Signing Service

The final stage of the Binance hack took just over two hours. Attackers logged into the third engineer’s computer and used the credentials stolen from the DevOps engineer. They deployed a container into the wallet infrastructure and launched a malicious script. From there they reached the internal wallet signing service and began moving funds outside the normal withdrawal process.

Money left in XRP, SOL, OP and BTC from the exchange’s hot wallets. Most of the value moved in roughly twenty minutes across ten blockchains. The attackers also abused EIP-7702, the Ethereum delegation feature introduced in the Pectra upgrade. It lets a standard account temporarily take on smart contract logic, which gave them on-chain authority over the wallet.

Multi-party computation complicated the response to the Binance hack rather than preventing it. No single key existed that the exchange could use to revoke the delegation immediately, so containment took longer. No wallet private key leaked at any stage, and the fraudulent transfers carried valid signatures throughout. Attackers who reach the authorization layer never need to break the cryptography underneath it.

Where Binance Stands After the Hack

The exchange has replenished roughly 98% of the affected tokens with company funds. About $2.85 billion sits frozen through cooperation with blockchain partners and other exchanges. Deposits and withdrawals have resumed after a temporary suspension, and Binance has committed to covering every customer loss from the hack.

The remediation programme has run wide across the affected environment. Binance rebuilt the wallet system in a new environment and replaced the compromised devices. It decommissioned the credentials and tokens tied to all three engineers, then ran enhanced scans across employee devices. The exchange has also added AI-powered anomaly detection and commissioned an independent external review.

North Korea has rejected the FBI’s attribution of the hack. The bureau is working with the Secret Service and IRS Criminal Investigation on the recovery effort. It has asked virtual asset service providers to block transactions touching the laundering addresses. Investigators expect the funds to move through decentralised finance platforms and swap services over the coming months.

What the Attack Chain Teaches

The most uncomfortable lesson in the Binance hack sits in June rather than October. The exchange detected the first compromise, replaced the device and rotated credentials. It still lost $6.34 billion five months later, because remediating an endpoint does nothing about the knowledge an attacker already took. Incident response that stops at the compromised machine leaves the attacker’s map intact.

The rest generalises to any company with engineers, developer tooling and internal automation. Recruitment now functions as an attack surface, and it reaches the technical staff whose workstations carry the most access. Developer tool installs deserve the same scrutiny as any other part of the software supply chain. Internal AI systems with broad reach need access boundaries that assume compromise rather than trust.

None of the controls that failed at Binance were exotic, and the hack required no novel technique beyond the zero-day payload. Patience, research and a convincing job offer did most of the work. Every organisation with developers hiring through public platforms faces the same opening move.