How exchange hacks happen

Pooled custody changes the threat model

An exchange normally keeps some assets available for routine withdrawals in internet-connected hot wallets and protects larger reserves with slower, more isolated signing arrangements. The exact labels vary, but the trade-off is stable: online availability improves service while increasing remote attack exposure; deeper isolation reduces exposure while making legitimate movement slower and operationally demanding.

Security depends on more than where a key is stored. Withdrawal software decides which requests reach signers, risk engines flag unusual destinations, staff approve exceptions, and accounting systems reconcile internal liabilities with blockchain assets. An attacker who influences enough of that pipeline may produce an apparently valid request without directly extracting raw private-key material.

Attackers move through people and systems

Initial access can come from targeted phishing, reused employee credentials, malware, vulnerable internet services, compromised vendors, or abuse by an insider. From there, attackers seek higher privileges, map signing workflows, disable alerts, and wait for a useful operational window. Mature attacks are often sequences of ordinary-looking actions rather than one dramatic cryptographic break.

Customer account takeovers are a related but distinct category. If a criminal controls one user's email, session, or recovery process, they may withdraw that user's balance while the exchange's custody keys remain secure. Users should distinguish platform-wide wallet compromise from individual account compromise because the evidence, containment steps, and likely scope differ.

Bybit's published timeline for its February 2025 incident says attackers spoofed the Safe user interface during a routine transfer from an Ethereum cold wallet, deceiving multisignature participants into approving changed wallet logic. The case shows why labels such as cold storage and multisignature do not settle the control question: signers need an independent way to verify the actual transaction and contract change presented for authorization.

Controls should constrain both keys and flow

A safer custody design uses independent signers, threshold authorization, least-privilege access, hardened build systems, withdrawal limits, destination controls, delayed high-risk changes, and continuous reconciliation. No control is enough alone. A multisignature policy offers limited protection if all signers run on the same compromised network or if employees routinely approve requests they cannot independently validate.

Monitoring should focus on behaviors that precede or accompany loss: unusual wallet consolidation, rapid reserve movements, new withdrawal infrastructure, disabled status pages, prolonged withdrawal delays, or inconsistent public explanations. These signals do not prove a hack; maintenance, migrations, and congestion can look similar. They justify reducing exposure and seeking reliable information rather than spreading an unverified claim.

Users manage duration and concentration

Before depositing, understand the legal entity, supported withdrawal networks, authentication and recovery options, allowlisting features, incident history, and whether asset or liability disclosures receive independent examination. Proof-of-reserves can show selected onchain assets, but without credible liabilities and control information it cannot establish solvency or show that assets are unencumbered.

Match custody duration to purpose. Funds needed for active trading may require exchange access, while assets with no near-term exchange purpose create uncompensated counterparty exposure if left there. Self-custody replaces exchange risk with key-management risk, so the correct choice depends on the user's ability, amount, transaction needs, and tested recovery process rather than a slogan.