A wallet can produce a cryptographically valid signature for a transaction its owner never intended to approve. That distinction is central to understanding the October 29 Binance theft—and to designing custody systems that remain safe when an internal service is compromised.
Our October 30 assessment identified abuse of the withdrawal and signing workflow as the leading explanation. The public evidence supports that interpretation, but it does not yet establish Binance's complete internal failure path. This post explains the mechanism and updates the original assessment with information reviewed on November 1, 2026.
What is confirmed, and what remains an assessment?
Binance's preliminary account put the impact at approximately $6.3 billion. It identified a compromised DevOps laptop and access to a service container as a possible route to unauthorized withdrawals, and said it had found no evidence of wallet private-key leakage. The same announcement reported approximately $3 billion frozen and committed SAFU and company funds to covering losses. Its updated notice clarified that deposits continued operating; withdrawals were suspended. These are Binance's statements, not independent verification of its internal forensics or balance sheet. Binance incident update.
Two developments matter for readers of our earlier report:
- Attribution has advanced. On October 31, the FBI attributed the theft to North Korea's TraderTraitor activity. Our October 30 report described DPRK involvement as an analytic hypothesis; it is now an official US attribution. That does not independently prove the internal intrusion sequence. FBI alert.
- Recovery is progressing in phases. Binance's November 1 recovery notice announced a phased withdrawal restart and said a post-incident report would follow completion of its investigation. Reopening services should not be confused with publishing a verified root cause. Binance recovery update.
How could a DevOps laptop reach a hot wallet?
The useful question is which trust boundaries an attacker could cross after acquiring an employee's access.
In a typical custody architecture, several systems sit between a workstation and a blockchain: identity services, production workloads, withdrawal queues, transaction builders, approval policies, signing infrastructure, and broadcasters. A hardware security module or multiparty computation system may protect key material while accepting requests from other services.
If an attacker can make a trusted service request signatures on malicious payloads, the signer may execute an unauthorized business action without exporting its key. Protecting key secrecy and protecting signing intent are separate requirements.
The following is our architecture-level reconstruction, not a disclosed diagram of Binance's infrastructure:
| Boundary | Possible failure | Evidence needed to establish it |
|---|---|---|
| Employee endpoint → production identity | Stolen credentials or session access permits entry to an internal workload. | Endpoint and identity-provider logs. |
| Production workload → withdrawal service | A compromised service submits attacker-selected withdrawal instructions. | Workload identities, API logs, and queue messages. |
| Withdrawal service → signer | Policy checks or approvals accept malicious transaction payloads. | Approval records, policy decisions, and signer audit logs. |
| Signer → blockchain | A valid signature is broadcast and accepted. | Public transaction data and resulting state changes. |
Binance's preliminary statement addresses the first part of that model. Public transactions expose the last part. The middle—how a service gained sufficient signing authority—is the critical unresolved question.
TRM's analysis points to signing-system abuse because wallets across different networks produced valid transfers for the attacker. An Ethereum-only explanation cannot account for that wider pattern. This supports our assessment of a shared authorization or orchestration failure; it does not identify the exact component that failed. TRM technical analysis.
What the Ethereum “batch transaction” actually did
The batch was a set of signed authorizations, not simply a list of withdrawals.
Range reports that, at 20:20:59 UTC on October 29, a transaction carried 50 pre-signed authorizations for the wallet labeled Binance 14. They covered different nonces—the account counters used to validate authorizations. This widened the window in which an authorization could match the wallet's changing state; it was not an unconditional guarantee of success. A subsequent sweep occurred three minutes and 24 seconds later. Range reconstruction.
EIP-7702 lets an account authorize a code delegation. Each authorization includes a chain identifier, delegate address, nonce, and signature. Ethereum validates those fields before installing the delegation. Once installed, calls to the account can execute the delegated code in that account's context. The delegation persists beyond the transaction that established it and can be cleared through a valid authorization to the zero address. EIP-7702 specification.
In this incident, the delegated code reportedly enabled repeated sweeps of incoming funds until revocation around 00:10 UTC on October 30. This turned a successful authorization into a continuing capability. Range reconstruction.
The distinction matters: the delegation explains Ethereum persistence; it does not explain how the attacker obtained Binance's valid authorizations in the first place. Calling the entire incident an EIP-7702 vulnerability would skip the upstream failure and leave the other affected chains unexplained.
What can the blockchain actually prove?
Public chain data can show an authorization, its signer, the installed delegation, transfers, subsequent calls, and state changes. It can establish what an account authorized and what execution followed.
It cannot, by itself, show whether a browser session was stolen, which internal API accepted a request, what an approver saw, or whether a policy engine was bypassed. Nor does a valid signature distinguish extracted keys from misuse of a legitimate signer.
This is why our conclusion remains qualified: the observed behavior is consistent with compromise of the signing workflow, while the precise intrusion path requires internal forensic evidence.
Frozen funds are not recovered funds
Elliptic estimated $6.43 billion stolen at its valuation and reported roughly $2.84 billion frozen at its analysis cutoff. Differences from Binance's figures reflect distinct snapshots and methodologies; neither number should be presented as a final net loss. Elliptic analysis.
An issuer freeze can prevent movement without returning assets. Recovery may need additional technical and legal steps. Native assets also lack the issuer-operated blacklist controls available on some stablecoins. A useful recovery account therefore separates gross outflows, immobilized funds, returned assets, and replacement capital.
The custody design lesson
Our engineering takeaway is to test the entire path from a withdrawal request to a signature, including what happens when an upstream service is hostile.
- Enforce transaction policy independently. A trusted workload identity should not, by itself, authorize an arbitrary destination, amount, or account-control change.
- Treat delegation as a privileged operation. Review the execution authority it grants, not only the immediate transfer value displayed to an approver.
- Limit shared exposure. Separate chain and wallet permissions, constrain hot-wallet inventory, and apply limits that remain effective after a service compromise.
- Practice containment under compromise. Exercise revocation, deposit routing, signer isolation, and partner coordination before an incident requires them.
The most valuable eventual post-mortem will explain which boundary failed, why independent controls did not stop the request, and what evidence demonstrates that the repaired system now rejects the same attack path.
Method: adapted from UBC's October 30 incident assessment and updated against the linked public sources on November 1. This is an independent analysis, not Binance's forensic report. We did not inspect Binance's internal systems or independently replay every cited transaction.