The $6.3 Billion Laptop: Binance’s Hack and the AI Question

Billions gone, customer funds “completely safe.” That was the opening chapter of the Binance hack. On the evening of October 29, observers spotted roughly $3.5 billion leaving the exchange’s Ethereum hot wallets in an abnormal burst of withdrawals. The figure soon grew: Binance put the assets affected by the incident at about $6.3 billion. It suspended deposits and withdrawals and promised to cover the losses.

Blockchain analysts arrived at different totals but saw much the same pattern: funds moving rapidly across multiple networks to new addresses. TRM Labs counted approximately $6.36 billion across ten blockchains, with most of it transferred in about 20 minutes starting at 20:05 UTC. Chainalysis estimated the stolen assets’ value at the time of the transactions at $5.912 billion. The initial $3.5 billion figure covered only Ethereum. By TRM’s reckoning, this was the largest crypto theft on record—roughly four times the size of the February 2025 Bybit hack.

The detail Binance may find hardest to explain sits off-chain: an employee’s laptop. According to the exchange’s preliminary investigation, an attacker may have compromised a DevOps engineer’s computer, used it to access a service container and submitted unauthorized withdrawal transactions. Binance said it had found no evidence that wallet private keys had leaked; PCMag reported its account. A multibillion-dollar crypto operation may have been breached through one member of staff’s work computer.

The transaction trail offers a closer look at the mechanics. CryptoTicker’s analysis describes a transaction containing 50 EIP-7702 authorizations, all pointing to a single attacker-controlled contract. The Ethereum feature lets an address delegate code execution, and Blockaid had earlier suggested it might have been used in the attack. That evidence helps reconstruct how transactions were carried out; it does not yet establish how the attacker first got in. TRM attributed the theft to North Korea’s TraderTraitor group using blockchain analysis and other intelligence. Its attribution and Binance’s preliminary account of the compromised laptop address separate parts of the investigation.

AI is a reasonable line of inquiry. Our earlier report on OpenAI’s agents examined traces in public logs of attempts to bypass access restrictions. Claims about models taking part in such operations can be tested against evidence of what they actually did. So far, the published analyses of the Binance incident offer no technical evidence that the attackers used AI.

“Binance's initial investigation suggests that the security incident may involve a compromised DevOps engineer’s laptop, which allowed an attacker to gain access to a service container and submit unauthorized withdrawal transactions.” — Binance, via PCMag

Blockaid did report a role for its models: detecting the unusual withdrawals shortly after they began. Speaking to Reuters, Trace Finance CEO Bernardo Brites raised the broader risk that AI could find vulnerabilities faster than defenders could fix them. He offered no evidence linking models to this particular attack. AI therefore appears in the reporting as a detection tool and a wider security concern. The claim that it helped carry out the hack remains unsubstantiated.

Containing the damage required some decidedly centralized intervention. Tether, Circle and other industry participants helped freeze assets, and Binance reported that roughly $3 billion had been blocked. Those funds still have to be recovered, so subtracting the frozen amount from the attack’s total and calling the remainder the final loss would be premature. Binance co-founder Yi He pledged that the exchange would cover all hacking losses. For customers, that promise now matters more than any sales pitch about blockchain transparency.

The blockchain gave observers a quick view of where the billions were going. Stopping them from leaving was the job of the exchange’s internal access controls, transaction checks and signing systems. Binance still has to explain which safeguards the attacker got past—and why they failed to stop withdrawals across ten networks. If its preliminary account holds up, the engineer’s laptop will become a very expensive reminder of how much can depend on the access privileges of a single work computer.