The Laptop Was Trusted. The Blast Radius Was $6.3 Billion.
Binance now says fake interviews, poisoned developer tools and compromised internal AI tooling helped attackers reach three engineers before $6.34 billion in assets was taken. The account still awaits independent review; frozen is not recovered.
AI-generated conceptual editorial illustration. It is not documentary evidence of the Binance incident or any identified employee.
Update, Nov. 3: the missing attack path
Binance now says the October 29 theft followed a multi-stage campaign involving a tampered developer tool, a fake job interview, malicious code disguised as an assignment, compromised employee endpoints and an internal AI tool. The company says attackers reached three engineers in stages before using stolen credentials and a deployment token against wallet systems.[1][3]
The brochure version of AI security usually skips the part where the internal AI system becomes one more route through a compromised workstation. Binance does not say the tool attacked anything on its own. It says people already inside the endpoint used it as a springboard.
Five wallets, four minutes
The report says 233.46 million XRP left five hot wallets for an attacker-controlled address in four minutes. Other assets followed. Binance says its response was slowed by the authorization method and the need to build new revocation procedures; the final unauthorized transfer settled nearly four hours after the first theft.[1][3]
Four minutes is enough time for an alert to become a history lesson. The control question is not whether monitoring noticed a large transfer. It is which independent boundary was supposed to stop the next signing transaction after it did.
Frozen, replenished and returned are different receipts
Binance says about $6.34 billion in assets was affected and about $2.85 billion has been frozen or intercepted, subject to reconciliation and price changes. It also says it set aside more than 100% of the money needed to cover the loss and that 98% of affected token assets had been replenished by October 31, with less-liquid assets still being purchased.[1]
Those are company statements. Replenishing reserves is not the same event as every customer receiving a completed reimbursement, and freezing assets is not the same as recovering or returning them. Keep the verbs attached to their actual receipts.
Attribution is official, not uncontested
The FBI attributes the theft to North Korea and calls the activity TraderTraitor. Reuters reports that North Korea rejects the allegation. Binance cites the FBI attribution in its report; it does not present its own independent public attribution evidence.[2][3]
Binance says it rebuilt the affected wallet system and internal AI tool, replaced devices, decommissioned related credentials and tightened wallet-service and device controls. It says independent security firms will review the incident, while some technical details will remain confidential.[1]
What still needs an answer
The November 3 report is far more specific than the first company update. It is still Binance’s account, not the promised independent review. The public record does not yet show which approval boundaries failed, why response procedures had to be built during the incident, or how much of the frozen value will ultimately be returned.
If a compromised workstation can reach wallet signing through an AI tool and a deployment token, which control was supposed to stop the fourth minute?
CORRECTIONS & UPDATES.
Update · Updated November 3, 2026: added Binance’s detailed attack path, the FBI attribution, the four-minute XRP withdrawal and current frozen-asset and reserve-replenishment figures. The article keeps company findings, official attribution and completed recovery separate.
FOLLOW THE SOURCES.
Checked Nov 3, 2026. Numbered references point to the evidence behind factual passages. Commentary is our own.
President Lee ordered officials to consider a government-wide security body after repeated attacks exposed the limits of inspections and post-incident response. The agency does not exist yet.
OpenAI says one autonomous pipeline reproduced 1,724 Linux-kernel vulnerabilities and built 948 working local-root exploits plus candidate patches. Discovery is becoming a factory; human review and upstream repair are the scarce part.