The Binance attack: 18 hours of preparation, three minutes to respond
The most revealing part of the Binance attack may be what happened before the money moved.
Three minutes and 24 seconds later, a single transaction removed approximately 167,464 ETH and balances in 63 tokens. PublicAML’s reconstruction documents the preparation and subsequent drain.
That interval raises a security question with implications far beyond cryptocurrency: can your organisation recognise a dangerous change in authority and act before that authority is used? Binance had in place some of the most advanced monitoring capabilities, and yet the attack was still a success.
The latest position
As of today (1 November), Binance has resumed bitcoin, ether and some stablecoin withdrawals following the $6.3 billion heist. They say that they have replaced 95% of the stolen assets using company funds. The FBI has agreed with community assessments and has called out North Korea’s TraderTraitor group as being behind the attack.
The Binance 14 drain was one part of a wider incident spanning multiple blockchains. Its short execution window (< 4 minutes) should not be confused with the duration of the entire attack.
Binance’s own investigation is ongoing. In their 30 October update, they stated that their initial findings suggest that a compromised laptop was a major factor in the attack, that additional breach points had subsequently been identified, and that it had found no evidence of leaked wallet private keys. A complete explanation of how the attacker obtained signing authority remains outstanding.
Preparation was visible before the target was obvious
The published Ethereum reconstruction describes a sequence of increasingly realistic tests.
| Reported activity | |
|---|---|---|
| 02:28 | The operator deploys an initial sweeper contract | |
| 13:51 | Two fresh wallets receive EIP-7702 delegations | |
| 14:10-14:24 | The operator calls the delegated wallets | |
| 15:59 | A second sweeper contract is deployed | |
| From 16:20 | A third wallet is used to test the second version with real funds | |
| 18:20 | A funding address sends approximately 1.08ETH to the operator | |
| 20:20:59 | Binance 14 receives a transaction containing 50 signed authorisations | |
| 20:24:23 | The first major sweep executes |
There were approximately 18 hours between the first contract deployment and the Binance 14 delegation. That establishes a visible preparation period; it does not, however, establish when the broader intrusion began.
It also does not mean the rehearsals would have been obvious warnings at the time. Fresh wallets testing delegations have little inherent connection to a particular exchange. Their significance oinly really becomes clearer once investigators can trace backwards from the victim. Then the penny drops!
Therein lies the difficulty of prevention: identifying which apparently ordinary events are becoming part of a consequential attack.
What EIP-7702 changed
EIP-7702 allows an ordinary Ethereum account to delegate execution to smart-contract code through a signed authorisation. It supports legitimate features such as transaction batching and sponsored transactions.
That capability also makes the choice of delegated code a critical security decision. A valid authorisation can cause code selected by an attacker to execute in the account’s context. The delegation persists until it is changed or cleared.
For Binance 14, the reported transaction carried 50 signed authorisations covering consecutive account nonces. For those that don't know, a nonce is a counter used when validating account activity. The batch provided a range of possible counter values rather than relying on a single prediction.
One crucial technical detail matters when interpreting the rehearsals: EIP-7702 increments the account nonce after each successful authorisation. Multiple consecutive authorisations can therefore all be valid as the list is processed. Nonce mismatches are skipped individually.
Consequently, sending two or three consecutive authorisations from a fresh account does not, by itself, prove that the operator was testing invalid nonces. The broader sequence supports deliberate preparation, but the operator’s precise intent at that point remains an inference.
The decisive signal was a change in authority
The strongest defensive signal was the moment a valuable production wallet acquired an unexpected delegation.
The Binance 14 delegation at 20:20:59 UTC and the sweep at 20:24:23. That left three minutes and 24 seconds between the recorded authority change and its first major use.
Why an unapproved delegation to recently deployed code didn't trigger immediate escalation is currently unknown, especially at these levels. Contract age and unavailable verified source code add context, and although neither alone establishes malicious intent, together they should raise an alarm.
A practical policy here should be pretty straightforward: production wallets may delegate only to explicitly approved code, through an approved signing workflow.
The implementation requires more than an alert rule. It needs ownership of the wallet inventory, control over what signing systems will approve, continuous monitoring of delegation changes, and a response procedure that can operate within minutes.
Prevention at the signing boundary can be particularly valuable. Because once a malicious delegation has been confirmed 'on-chain', defenders are already responding to an active compromise.
Detection must lead to a usable response
It is tempting to conclude that nobody noticed the delegation. Public blockchain data won't be able to establish that.
An alert may well have been triggered, delayed, misclassified or escalated without an effective response. Finding out what happened inside Binance requires internal telemetry and the eventual incident report. Whether or jnot those details will ever be made completely public is another matter.
The only defensible conclusion at this time is that the delegation was followed by a successful sweep within a very short interval.
Lessons to take from this attack
The lessons here extend beyond crypto wallets
The same pattern appears when an application receives excessive OAuth permissions, a cloud identity gains administrative access, or a supplier changes the bank details attached to an invoice.
The harmful transaction can look valid because the attacker has already changed who or what is trusted.
Security controls should therefore examine changes in authority alongside the actions that follow them. A newly introduced source or destination, permission or execution path deserves assessment before it becomes an accepted part of normal operations. All to often this is not the case.
The Binance timeline illustrates how small that opportunity can be and how devastating an effect it can have. Preparation can unfold for hours, while the clearest actionable signal arrives only minutes before the loss.
The value of threat intelligence depends on whether it reaches a decision point in time, and whether someone, or something, can act on it.
Thanks to PublicAML and Hacken for their rapid analysis and insights into this heist.