A financial institution can process a transaction using valid authentication and a genuine digital signature even when the payment was not authorised for a legitimate business purpose. This can occur when attackers compromise systems responsible for initiating or approving transactions rather than the cryptographic credentials themselves.

The 29 October 2026 cyberattack on Binance, a major cryptocurrency exchange, involved approximately $6.34 billion in affected digital assets, according to the company. Binance reported that attackers reached its wallet-signing infrastructure without evidence of private-key theft. Independent blockchain investigations identified large unauthorised transfers, although the internal attack sequence remains subject to investigation.

For banks developing digital asset custody, tokenised deposits and programmable payments, the incident raises questions about financial authority, transaction controls and the time available to contain fraudulent transfers.

Attackers moved from engineering access to transaction authority

Binance's preliminary findings, published on 3 November, describe a campaign beginning in June with a compromised engineering workstation and a separate compromise in September. The company reported that attackers subsequently used an internal AI tool to reach other wallet engineers' systems and obtained credentials permitting software deployment into production.

On 29 October, attackers allegedly used this access to execute malicious code that reached the wallet-signing service, enabling transactions outside the normal withdrawal process.

The role of the internal AI tool remains unclear. The reported sequence indicates a possible failure in separating development activities from financial operations. Software deployment privileges should not ordinarily confer authority to approve asset transfers. Whether Binance maintained that separation, and how attackers crossed it, remains subject to independent review.

TRM Labs, a blockchain intelligence company, estimated that approximately $6.36 billion moved across ten blockchains, most within about 20 minutes. Elliptic, a blockchain analytics firm, calculated approximately $6.43 billion using a different methodology. These assessments establish the scale and speed of transfers but not the internal attack sequence.

Earlier breaches exposed weaknesses in transaction authorisation

In 2016, attackers compromised Bangladesh Bank's international payment systems and submitted fraudulent instructions through the Society for Worldwide Interbank Financial Telecommunication (SWIFT). They attempted to transfer nearly $1 billion from the central bank's account at the Federal Reserve Bank of New York. Approximately $81 million was transferred. The New York Fed stated that its systems were not compromised and the instructions had been properly authenticated through SWIFT.

The February 2025 attack on cryptocurrency exchange Bybit involved a compromised transaction interface supplied by Safe, a digital asset wallet infrastructure provider. Bybit personnel believed they were approving a routine transfer, while the underlying instruction changed the wallet's operating logic and enabled the theft of approximately $1.5 billion.

The incidents involved different points of compromise: Bangladesh Bank's payment instructions, the transaction information presented to Bybit's signatories and Binance's infrastructure requesting transaction signatures. In each case, legitimate systems processed instructions that did not reflect the institution's intended transaction.

Cryptographic protection cannot establish payment intent

Multi-party computation (MPC), used in digital asset custody, distributes cryptographic signing authority across separate components. Rather than storing a complete private key in one location, the system generates a valid signature when the required participants complete the signing process, reducing dependence on a single device or key component.

However, MPC cannot independently establish legitimate payment intent. If attackers control an application permitted to request signatures, the infrastructure may sign fraudulent transactions without exposing the private key.

Privileged-access management and credential isolation restrict engineering access but may be less effective when attackers obtain authorised sessions or deployment credentials. Banks and custodians can reduce exposure by independently verifying transaction destinations, values and purposes before signing, provided those approval rules cannot be modified through compromised access.

AI-enabled operations introduce related risks. Internal AI tools may access information, modify software or interact with applications through assigned permissions. The evidence does not establish that an AI model autonomously caused the Binance breach, although the permissions available through its internal tool remain under investigation.

Lee Wan Sie, Cluster Director of AI Governance and Safety at Singapore's Infocomm Media Development Authority (IMDA), discussed defining what AI agents may access, decide and execute without further approval. She also highlighted cumulative financial exposure when automated systems conduct numerous transactions within individual approval limits. Similar considerations apply to digital asset wallets operating across networks.

Detection must restrict transactions before settlement

Binance's preliminary chronology identifies potentially suspicious activities before the transfers, including access between engineering systems, privileged deployments and signing requests. Public disclosures do not establish which alerts were generated or whether monitoring controls failed.

Individually, these activities may appear authorised. Detecting misuse may require correlating engineering activity, deployment records and transaction requests across operational and financial systems.

Marlon P. Sorongon, Chief Information Security Officer at Maybank Philippines, described how unusual device activity, transaction patterns and location changes can trigger additional verification. Although primarily addressing customer fraud, these measures demonstrate the value of transaction context beyond identity authentication.

Bybit reported intercepting more than 30,000 suspicious withdrawal requests during the first half of 2026, representing over $700 million in estimated potential customer losses. Bybit reported an average initial risk review time of 4.7 minutes. These company-reported figures do not independently establish losses prevented.

For banks and custodians, detection must trigger enforceable restrictions before settlement. Monitoring unusual destinations, permission changes and cumulative transfers can support automated transaction limits or additional approvals. With most stolen assets reportedly moving within about 20 minutes, intervention must operate quickly and independently of potentially compromised systems. Alerts after settlement may support investigations but cannot necessarily prevent losses.

Suspending withdrawals did not revoke existing authority

The limits of containment emerged during Binance's response on Ethereum, a blockchain supporting programmable transactions. Attackers used Ethereum Improvement Proposal 7702 (EIP-7702), which allows an account to delegate certain transaction-execution powers to software operating on the blockchain.

According to Binance, attackers established permissions allowing malicious code to continue moving assets independently of its normal withdrawal system. Suspending withdrawals did not revoke these permissions because they remained active on-chain.

MPC complicated emergency revocation because no single private key was available to remove the permissions immediately. Binance's security team needed a separate procedure through its distributed signing infrastructure, prolonging the containment process.

The last unauthorised transfer was confirmed at 23:56 UTC on 29 October, while the malicious delegation was revoked at 00:10 UTC on 30 October. This 14-minute interval does not represent the full containment period: transfers had begun several hours earlier, and the final transaction had already been submitted before confirmation.

The incident distinguishes suspending a payment application from removing authority already established on an external network. A kill switch that stops withdrawals but cannot revoke signing permissions provides only partial containment. Banks developing programmable payments need tested emergency procedures that work even when ordinary infrastructure is compromised.

Binance reported approximately $2.85 billion in affected assets frozen or intercepted through coordination with issuers and other parties. Intervention mechanisms differed by network. Some stablecoin issuers can freeze specified tokens, while Bitcoin and Ether generally lack equivalent centralised controls.

External coordination helped freeze or intercept some affected assets but could not immediately stop all transfers. Banks must establish how quickly compromised permissions can be revoked and what transactions remain executable after their primary payment systems are suspended.

Recovery requires separate technical and financial assessments

Samar Sen, Head of International Markets at Talos, a technology provider serving institutional digital asset markets, discussed the treasury and operational requirements of continuous settlement. Faster settlement can require earlier funding and leave less time to intervene in fraudulent transfers, linking custody security to liquidity management and customer obligations.

Binance stated on 3 November that it had rebuilt the affected wallet system and internal AI tool, replaced compromised devices, revoked credentials and strengthened monitoring. These measures address the reported attack path but do not independently establish that all attacker access was eliminated.

Remediation requires evidence that malicious changes no longer persist, compromised identities lack access and transaction controls remain enforceable. Binance announced an independent cybersecurity review, whose findings had not been published as at 10 November. Its rebuilt infrastructure therefore remained subject to external validation.

Operational resilience also requires maintaining essential services during disruption. The Monetary Authority of Singapore (MAS) has convened a technology resilience panel that recommended exercises assessing institutions' ability to operate during technology failures.

Financial recovery must be assessed separately. Binance pledged to absorb losses and reported that 98% of affected token assets had been replenished as at 31 October using company resources. Replenishing customer balances does not establish that stolen assets were recovered. Similarly, frozen or intercepted assets cannot automatically be treated as returned funds.

In its 1 November announcement, Binance described phased withdrawal restoration beginning with selected assets and networks after security checks. Deposits remained operational. Customers began regaining withdrawal access roughly two and a half days after the attack, although disclosures did not establish when every affected service was fully restored.

Absorbing losses, replenishing assets and progressively restoring services demonstrated financial support and operational recovery, but did not establish that security weaknesses had been eliminated.

Assessment therefore requires three separate measures: stolen-asset recovery, resources available to meet customer obligations and independent validation of rebuilt infrastructure. Binance is not a conventional deposit-taking bank, and its arrangements cannot be directly compared with bank capital, liquidity or deposit protection requirements. The common consideration is whether customer obligations can continue to be met after an operational loss.

Banks face decisions over the allocation of financial authority

The Bangladesh Bank, Bybit and Binance incidents involved compromised systems performing different functions within legitimate financial processes. Cryptographic protection, access controls and monitoring address particular risks, but their effectiveness depends on whether compromised systems can influence financial approval and execution.

For banks investing in digital asset custody, programmable settlement and AI-enabled operations, the decision concerns how much transaction authority to assign to connected applications and which safeguards should remain independent. This includes where transfer limits are enforced, which activities require separate approval and how compromised permissions can be revoked.

The Binance incident demonstrates why institutions must measure how quickly they can contain unauthorised transfers, restore essential services and independently verify remediation. These capabilities determine the financial exposure created by a compromise and the ability to maintain customer confidence during recovery.