Assessing Crypto Exchange Soundness: Large-Scale Hack Response and Institutional Operational Risk

At 20:05 UTC on October 29, the first outflow occurred from a Binance wallet on the XRP Ledger. According to TRM Labs, most of the funds left across 10 blockchain networks over about 20 minutes, and a second wave of outflows followed about 3 hours later^[7]. Binance formally put the total loss at about $6.3 billion^[1]. Chainalysis counted $6.4 billion, and Elliptic counted a total of $6.43 billion across 11 chains including NEAR^[8][4]. On October 31, the US Federal Bureau of Investigation (FBI) attributed the attack to TraderTraitor, a North Korea-linked hacking group^[9]. North Korea's Kim Yo Jong, for her part, formally denied any involvement on November 3^[10].

Table 1. Key timeline from the first outflow to the Phase 2 withdrawal reopening

※ The detailed timeline of outflows by chain and of freeze and coverage announcements is set out in Appendix ④.

The risk of a structure that combines operating access and transfer approval authority

According to blockchain security firm Blockaid, the attacker registered malicious EIP-7702 delegation code on the Ethereum hot wallet "Binance 14". Just 3 minutes 24 seconds later, a single transaction drained 167,464 ETH along with a number of other tokens^[13]. EIP-7702 is an Ethereum standard feature that uses a valid signature from the wallet owner to have the code of a designated smart contract executed directly in that wallet's context^[14]. From the bitcoin hot wallet, 18,270 BTC left through ordinary signed transactions. Citing the fact that multiple blockchain wallets produced valid withdrawal signatures within a short window of about 30 minutes, TRM Labs assessed that the core infrastructure the exchange uses to sign hot wallet transactions had likely been compromised^[7].

Binance said its initial investigation suggested that the attacker reached the software runtime environment (service containers) via the personal laptop of a development and operations (DevOps) engineer, and that unauthorized withdrawal requests may have been executed through that route^[1]. The company later said it had identified additional points of breach but had found no evidence that wallet private keys themselves were leaked. It plans to publish a detailed incident report once its full investigation is complete^[1].

INTERPRETATION If the signing and approval system itself could be abused without any private key being leaked, then the firm-wide internal control process running from withdrawal request to approval needs to be re-examined from the ground up, along with the physical custody of keys. If an attacker who breaks into operating systems can bypass and obtain final withdrawal approval authority as well, damage spreads in an instant across every system that shares the approval path, even when assets are physically spread across chains.

Recovery and coverage paths that diverged sharply by asset type

Table 2. Freeze mechanisms and coverage status by type of stolen asset

On a blockchain, a "freeze" ties up stolen assets so that they can no longer move, while "recovery" means the process by which the exchange actually gets those assets back. "Covering customer assets," by contrast, means the company putting in its own resources to replace them with assets of the same type, and a "customer payout" is complete only when customers actually withdraw their assets through withdrawal requests. Frozen amounts counted by TRM or Elliptic therefore do not translate directly into cash-equivalent amounts returned to the exchange. Tether said it had itself frozen more than $1.5 billion of USDT, and Reuters reported that Circle's USDC freezes exceeded $1.1 billion^[16][17].

INTERPRETATION Even when the same amount is stolen, the practical recovery path and its cost differ completely depending on which type of asset was taken. With stablecoins, central control by the issuer can quickly block an attacker from cashing out, but separate legal and administrative steps are needed before those funds can be returned to the exchange. Bitcoin and ether, on the other hand, have no function for forcibly locking an address, so before any recovery an exchange may have to deploy its own liquidity or buy the assets directly in the market to make customer balances whole. The remaining roughly 5% that Binance has yet to fully secure consists of tokens with very low liquidity, and the time needed to source them is also a factor delaying the point at which coverage of all customers is complete^[1].

Binance's recovery should be assessed along three separate dimensions: the share of stolen assets it has re-secured, the range of assets for which withdrawals are actually allowed, and its liquidity capacity to absorb withdrawal demand. The "95% coverage" the company announced publicly refers to the first of these: the share of total stolen assets that it has replenished from its own resources.

Table 3. Three key indicators for assessing an exchange's state of recovery

Assets secured, but withdrawal routes not yet open

Binance began withdrawal services at 06:00 UTC on November 1, starting with bitcoin, for the 15 networks designated for Phase 1^[1][12]. According to a status page checked by The Block the same day, however, withdrawals were still blocked for 531 of 552 coins in total^[18]. As of 9:32 a.m. KST on November 3, Binance's public status data showed withdrawals allowed for only 14 assets and 27 asset-network pairs. After Phase 2 withdrawals reopened at 1 p.m., withdrawable assets had expanded to only 27 assets on 14 networks, 40 pairs in total, at 1:43 p.m. The 10 major assets that had completed proof of reserves (PoR) verification could each be withdrawn on at least one network, but only 23 of their 72 network pairs in total were actually open for withdrawal (our estimate)^[6]. The same crypto asset is treated as a separate pair for each blockchain network it uses.

Of the 1,325 pairs on which withdrawals remained blocked, 505 were marked "Wallet Maintenance," 389 "Low Hot Wallet Balance" and 341 "Unsupported Token" (our estimate)^[6]. Binance explained the low hot wallet balance by saying that it was temporarily limiting withdrawals while it refilled and moved funds into its hot wallets. Further networks will open in sequence only after each network fully passes its own security and stability checks^[1].

INTERPRETATION Even when assets are fully recorded in the proof-of-reserves ledger, they can move only once the technical withdrawal route that lets customers actually send them out is opened. Safely refilling hot wallet balances and running technical checks network by network takes real time. Withdrawal delays should therefore not be read straight away as a fundamental asset shortfall at the exchange. Conversely, the limited suspension reasons the company has disclosed are not enough to accept without question that every withdrawal delay is purely a matter of operating procedure.

What SAFU absorbed and how company resources were brought in

SAFU (Secure Asset Fund for Users) is an emergency reserve that Binance built up by converting the entire fund, about $1 billion, into bitcoin in February^[19]. Just before the hack, the SAFU bitcoin wallet was worth about $1.2 billion, around 19% of the total loss in this incident (about $6.3 billion) (our estimate)^[2][3][1]. On October 30, Binance set aside more than $6.34 billion of company funds to cover customer assets, of which about $5 billion was already held as reserves in the same types of assets as those stolen. On October 31 it then formally designated part of SAFU, the surplus in its futures insurance fund above risk-management thresholds, and additional company resources as funding for loss coverage^[1]. The same day, 7,499 BTC, half of the reserve, was transferred out of the SAFU wallet^[2].

As of 9:31 a.m. KST on November 3, about 7,500 BTC remained at SAFU's public blockchain address, worth about $577 million at the bitcoin price at the time. At the same price, about $423 million of additional funds would be needed to reach the SAFU refill target of about $1 billion that Binance has stated (our estimate)^[2][3][1]. Before the incident, Binance's total public wallet balances amounted to about $144 billion, but because that figure includes assets deposited by customers worldwide, it must be kept strictly apart from the resources the company can freely use to cover losses^[20][21].

INTERPRETATION Since SAFU alone could not cover even 1/5 of the total loss, a large injection of additional company resources was unavoidable to cover customer assets in full. Binance added its own corporate funds to SAFU and part of its futures insurance fund. With SAFU's bitcoin holdings cut in half as a result, its capacity to absorb any further black swan (an unexpected incident) is substantially weakened until the fund is fully refilled to its target level. If another security incident or asset loss occurs during this vulnerable period, the company will bear an even heavier financial burden.

The limits of what proof of reserves and net outflow data show

At 12:45 UTC on November 2, Binance published proof of reserves for 10 major crypto assets as of 00:00 UTC on October 31. Customers holding BTC, ETH, USDT, BNB, SOL, USDC, XRP, USD1, RLUSD and ZEC can each verify whether their deposit balances were properly included in this attestation^[1]. According to crypto.news, reserve ratios for the major assets stood at 100.04% for bitcoin, 101.30% for BNB and 100.00% for Tether (USDT). The USDT reserve ratio in particular came down somewhat from 103.88% as of October 1, and the number of assets covered by the regular report was cut sharply from 55 to 10^[22]. Binance plans to widen the range of covered assets again on November 4 and update its balance records^[1]. Proof of reserves is a procedure that confirms holdings of the covered assets and the level of backing against customer balances at a particular snapshot. It is not a full audit by an accounting firm certifying the exchange's total contingent liabilities and overall financial soundness^[21].

Cumulative net outflows in the first 9 hours 30 minutes after withdrawal services partially reopened reached about $3.64 billion^[12]. A Binance spokesperson then told Reuters on November 2 that cumulative net outflows had reached $4.72 billion^[17]. According to DefiLlama data cited by The Block, 24-hour net outflows peaked at about $7 billion before easing to about $3.29 billion as of November 2^[23].

INTERPRETATION Net outflow data from a controlled state, in which withdrawals are allowed only for a limited set of assets and networks, make it hard to assess objectively whether the exchange could cope if all customers demanded full withdrawals. If the range of permitted withdrawals is opened up fully, pent-up withdrawal demand could surge and net outflows could spike again.

Table 4. Bitcoin and ether returns around the hack

※ Returns calculated from the close of the last 15-minute bar completed by each reference time. For example, the reference price for 20:05 is the close of the bar that ended at 20:00 sharp (our estimate)^[3].

In the 1 hour after the first outflow, bitcoin fell 2.48% and ether 3.36%. From the 20:00 bar to the 00:15 bar, just after the pledge to cover losses in full, the cumulative decline came to 3.72% for bitcoin and 3.96% for ether. Over the three days from the day of the incident, the maximum drawdown on a closing basis was 5.0% for bitcoin and 4.9% for ether, and on an intraday-low basis 6.2% and 7.9% respectively. Both assets showed only minimal volatility around the Phase 1 withdrawal reopening (our estimate)^[3].

INTERPRETATION Because bitcoin and ether fell together right after the first outflow, it is hard to read the move simply as a shock from bad news specific to bitcoin. Ether, however, was also a direct target of this hack, and the FBI said the attacker was converting other stolen altcoins into ether^[9]. A simple comparison of the two assets' returns therefore cannot cleanly separate a bitcoin-specific price shock from market-wide risk-off sentiment.

The price decline and its transmission through supply, liquidations and risk-off

According to Elliptic's count at 07:00 UTC on November 2, the 18,270 BTC taken in the initial theft were split across 19 separate blockchain addresses and have sat there without further movement. However, in the course of converting other stolen funds into bitcoin, funds worth about 28 BTC were found to have passed through CoinJoin, a technique that bundles many transfers into one to obscure tracing^[4].

A heavy wave of forced liquidations hit the market early in the incident. According to CoinGlass data cited by BeInCrypto, $627.7 million of derivatives positions were liquidated in just 1 hour after news of the hack reached the market, and about 97% of that volume was long positions betting on price rises^[5]. Bitcoin was already in a downtrend, having fallen 3.3% over the week and 2.6% over the month to the day before the incident (our estimate)^[3].

Trading volumes also surged in the domestic Korean crypto market right after the incident. Daily trading value in Upbit's BTC/KRW market reached KRW 168 billion on October 30 (UTC daily bar), 2.49 times the average for the prior 7 days (October 22–28). It then returned to the pre-incident 7-day average level on November 2 (our estimate)^[25].

INTERPRETATION No large further movement of the stolen bitcoin was observed, yet the sharp market-wide price decline and large derivatives liquidations were real. Direct market selling pressure from the stolen bitcoin alone is therefore not enough to explain the initial price shock. How far cascading liquidations amplified the decline, and how much risk-off sentiment from weaker investor confidence played a part, are hard to quantify from these data alone. Separately, the minimal price movement observed right after withdrawals reopened may also reflect recovery expectations being priced in ahead of time, so it cannot be equated with a full assessment of the exchange's ability to pay.

A multi-angle comparison with 8 past major exchange incidents

The comparison covers 8 past exchange incidents in total. Alongside the Mt. Gox collapse, which saw the largest bitcoin loss on record, we selected the main cases that dealt a major shock to the industry^[26]. Each incident falls into one of three types, depending on who (customers or the exchange) absorbed the loss and how. Mt. Gox and FTX halted withdrawals and then entered bankruptcy and rehabilitation proceedings^[27][28]. Bitfinex chose a forced haircut, cutting customer account balances across the board by 36.067%^[29]. Coincheck, KuCoin, DMM Bitcoin, Bybit and Bitget, on the other hand, covered customer losses using recovered funds together with their own resources, insurance funds and similar sources.

Setting each incident's close on the day before its reference date at 100, we calculated how far bitcoin fell to its lowest close among the UTC daily bars over the three days (3D) and seven days (7D) starting on the reference date. The intraday maximum drawdown was calculated separately from the lowest intraday price over the three days^[3]. For Mt. Gox and FTX, the date withdrawals were fully halted was used as the reference date. In FTX's case, the hack itself occurred on November 12, after the withdrawal halt, so separate values based on the hack date are also shown^[30].

Table 5. Loss size and maximum bitcoin drawdown in past major exchange incidents, by type

※ Prices and shares of market cap are based on raw data from Bitstamp and Blockchain.com^[3][38]. The Mt. Gox share uses circulating supply at the end of February 24, the day before the site closed; the others use the day before the hack was disclosed^[26]. ETH values are not shown for incidents before 2017, for which Bitstamp has no ETH prices. Binance's 7-day value can be calculated after the UTC daily bar closes on November 4. 0.0% covers both cases with no close below the reference price and cases with a small decline that rounds to 0.0% at one decimal place.

Table 6. Treatment of customer assets and time to withdrawal reopening or repayment in past incidents

※ Day counts are calendar-day differences between the withdrawal halt and each reopening, distribution or transfer date (our estimate). The time to reopening differs from the time to post-bankruptcy distributions or account transfers. Coincheck dates are as reported in Japan time.

INTERPRETATION The 3 past cases in which the three-day maximum drawdown on a closing basis from the reference date exceeded 10% were all catastrophic cases in which withdrawal services were never reopened or customer accounts were forcibly cut. By contrast, in the other 5 cases, where the exchange stepped in to cover losses in full, the three-day maximum drawdown on a closing basis was at most about 2.2%. Even among these well-handled cases, however, the time needed to normalize varied widely. KuCoin announced the reopening of deposits and withdrawals for all tokens after 57 days but kept some withdrawal limits and exceptions, while DMM Bitcoin gave up on reopening withdrawals itself and moved customer accounts to another firm after 281 days^[34][44]. Bybit responded without halting withdrawals, yet its 7-day maximum drawdown on a closing basis widened to 14.5%^[46]. That said, Bitfinex's haircut policy was announced after the three-day observation window, and FTX's 22.9% is the drawdown during the withdrawal halt before the hack^[29]. Short-term price reactions alone are therefore not enough to assess each exchange's loss-bearing structure and financial soundness on a uniform basis.

Binance's 3-day drawdown in this incident (5.0%) is the largest among the 6 exchange-covered cases compared here. Its loss as a share of bitcoin market cap (0.39%) was meaningfully larger than in the other covered cases (0.02–0.28%), and the hack came at a fragile point, with bitcoin already under 3.3% of downward pressure over the preceding week. In addition, the October 29 daily bar used as the reference date also includes about 20 hours of trading before the first outflow occurred (our estimate)^[7].

The loss relative to market size and to exchange resources

Binance's hack loss of $6.3 billion equals 0.39% of bitcoin's total market cap the day before the incident (about $1.6335 trillion) (our estimate)^[3][38][1]. That is about 1/14 of the share calculated for the Mt. Gox collapse on the revised estimate of 650,000 BTC (5.22%). Against the share based on the originally announced 850,000 BTC (6.83%), it is only about 1/18 (our estimate)^[26]. This ratio measures the loss against the size of the bitcoin market at the time; it does not directly indicate the actual price shock or the exchange's ability to pay.

To assess an exchange's real financial risk accurately, however, the comparison should be with the size of the resources the company can mobilize immediately to cover losses in an emergency, not with the total loss alone. In this incident the loss was more than 5 times the value of the Secure Asset Fund for Users (SAFU), and Binance said that, by mobilizing its own corporate funds and other resources, it had re-secured about 95% of the stolen assets within three days of the incident^[1].

The outflow of 18,270 BTC in this incident was accompanied by valid signatures from Binance wallets^[7]. The company said its official investigation had found no evidence that the wallet private keys themselves were stolen. In the investigation disclosed so far, the targets flagged as possibly compromised are not the bitcoin protocol itself but the exchange's internal operating environment (DevOps) and its signing and approval system^[1].

INTERPRETATION Ordinary companies and institutional investors that hold bitcoin as a large treasury asset should manage the internal approval framework and operational controls of the custodian holding it on their behalf as a core compliance risk, in addition to the market price risk of the asset itself. Losses arising from an exchange's operations, as in this Binance incident, must be clearly distinguished from a security failure of the blockchain network itself. Even with bitcoin on the balance sheet, without rigorous custody and withdrawal control procedures the asset cannot be put to use in time when it is needed in an emergency.

Table 7. Operational control criteria to check when using an exchange or custodian

Korean financial regulators' approach to supervising virtual asset service providers points the same way. On November 2, the Financial Services Commission held an emergency inspection meeting with the Financial Supervisory Service, the Financial Security Institute, the Digital Asset eXchange Alliance (DAXA) and the heads of major domestic exchanges. Administrators' system access controls, strict withdrawal approval procedures, cold wallet storage ratios and the status of insurance and emergency reserves were the core items under review. Regulators asked domestic exchanges to raise the share of assets held in cold wallets, build up their own emergency reserves and monitor deposits and withdrawals at all times^[47].

INTERPRETATION When assessing a crypto exchange or a specialist custodian, one should not be swayed by total asset holdings or by fragmentary proof-of-reserves figures alone. What needs to be verified, dispassionately, is whether it has in place in advance the substantial resources to return deposited assets to customers in full even after a major security incident, and transparent operating procedures to deploy them quickly. Even Binance could not fill the huge loss with its emergency fund SAFU alone and had to combine it at short notice with part of its futures insurance fund and its own corporate funds.

Domestic and overseas price gaps after adjusting for the USDT price

On October 30, Upbit and Bithumb posted notices urging caution when sending crypto assets to Binance and said they could restrict withdrawals if necessary^[48][49]. This was a milder step than in the Upbit hack in November 2025, when major domestic exchanges fully suspended withdrawals to Upbit^[50].

As of 9:29 a.m. KST on November 3, Upbit's KRW price for bitcoin was 1.15% higher than the Bitstamp USD price converted into won. The USDT premium on Upbit, calculated at the same exchange rate, was 0.99% (our estimate). At the previous observation points, the gaps were 1.6% and 1.3% respectively at 9:49 a.m. on October 30, and 0.52% and 0.39% respectively at 9:29 a.m. on November 2^[3][25][51].

Table 8. Domestic and overseas price gaps for Upbit crypto assets, converted at the USDT price

※ Formula: Upbit KRW price ÷ (Bitstamp USD price × Upbit USDT KRW price) − 1. Based on 1-minute bar closes for the same minute (our estimate)^[25][3].

Converted at the KRW price of USDT, the domestic and overseas price gaps for the 3 major assets, bitcoin, ether and XRP, narrowed to within 0.3% at all four observation points. Once the effect of the USDT price gap on Upbit is taken out of the domestic KRW premium, the residual price gap for each coin was minimal. The USDT price, however, reflects a combination of won-dollar exchange rate volatility and USDT supply and demand within the domestic market. Indeed, on October 30 total trading value in Upbit's KRW markets jumped to 2.49 times the daily average of the prior 7 days (October 22–28) for bitcoin, 2.31 times for ether and 1.88 times for XRP, before returning to the pre-incident 7-day average level on November 2 (our estimate)^[25].

INTERPRETATION Although Upbit and Bithumb issued caution notices warning that withdrawals to Binance might be restricted, domestic and overseas price gaps after USDT conversion were not large at the 4 observation points analysed. This contrasts clearly with the halt of deposits and withdrawals at Upbit in November 2025, when the price of Jito (JTO) was more than 35% above overseas prices^[52]. Still, the limited sample used here does not allow us to generalize that asset movement between domestic and overseas markets was fully smooth, or that the domestic-overseas price gaps came entirely from exchange rate moves.

Within three days of the incident, Binance announced a plan for funding the coverage, said it had re-secured about 95% of the stolen amount in the same assets, and reopened withdrawal services for some major assets in phases. In the past, Mt. Gox and FTX failed to overcome their withdrawal paralysis and went straight into court-supervised rehabilitation and bankruptcy proceedings. It took Mt. Gox about 10 years to begin BTC/BCH repayments and FTX about 2 years to make its first cash distributions to small creditors.

Table 9. 5 key conditions for judging whether a crypto exchange crisis is subsiding

INTERPRETATION The first thing to monitor is how fast the number of asset-network pairs with normal withdrawals expands and how smoothly customer withdrawals are actually processed. If the exchange buys the remaining tokens, widens the set of permitted withdrawals and still handles the incoming withdrawal requests without delay, its customer payout system can be judged to have effectively returned to a normal track. Conversely, if processing delays arise or withdrawal suspensions caused by low hot wallet balances recur, the exchange's liquidity management capability needs to be fully reassessed.

INTERPRETATION The follow-up reserve count scheduled for November 4 is the next piece of evidence for confirming the scope of the assets covered. Whether the stolen bitcoin flows to centralized exchanges or OTC prime brokerage desks also needs to be tracked separately, to gauge the risk of potential selling supply reaching the market. Whether SAFU is fully refilled, together with the official incident investigation report and follow-up measures, should be used to verify whether the exchange has restored the financial buffer to face the next crisis and whether it has improved its underlying security architecture.

① Limitations

• The comparison set is 8 past exchange incidents plus the Binance hack. In some cases the compensation policy was decided after the price observation window, and the sample is too small to estimate statistically the effect of the loss-bearing structure from differences in drawdowns.
• Price comparisons start from the withdrawal halt date for Mt. Gox and FTX and from the hack disclosure date for the other exchanges. The 3-day and 7-day windows are based on UTC calendar days, so the first day also includes trading before the incident occurred. Point-in-time returns are based on 15-minute bar closes and do not capture moves within 15 minutes.
• BTC, ETH and XRP dollar prices come from a single exchange, Bitstamp. Domestic prices are Upbit minute bars, and exchange rates are the daily published rates available at each observation time. The KRW premium does not deduct conversion costs or fees and so differs from actual arbitrage returns.
• Withdrawal access is the value obtained by querying Binance's public deposit and withdrawal status data at specific times. The denominator also includes pairs that were blocked for reasons unrelated to the incident, such as "Unsupported Token." Asset-level reserve ratios in the proof of reserves are cited from press tallies.
• Binance's estimate of about $6.3 billion and the estimates of the various analytics firms differ in the chains and assets included and in the valuation time. Frozen and seized amounts are kept separate from returned amounts, and the final net loss has not been calculated.

② Formulas for Our Own Estimates

The formulas, reference times and sources applied to our estimates in the main text are set out below.

③ Methodology

• Source priority: official materials from companies and regulators, then raw data and aggregator databases, then specialist media, then syndicated media, then community sources.
• Source classification: a source is classed as primary only when the original was checked directly. Content conveyed through reporting is attributed to the outlet that reported it.
• Reference and access dates: values fixed at a specific point in time carry their reference date; dashboard figures that change in real time carry their access date.
• Conflicting figures: where different figures are found, both are presented. Where a reference value has to be chosen, disclosure of the calculation scope, clarity of the query time and separability of line items take precedence over the outlet's profile, and the basis for the choice is stated in the main text.
• Our estimates: each of our estimates is presented individually in Appendix ② with its formula, reference date and limitations.
• Legal and regulatory interpretation: regulators' materials are cited only to explain the general classification of structures and their possible application, and are not read as an official classification or approval of any specific product. Descriptions of product structures and regulatory interpretation are kept in separate paragraphs.
• Conflict of interest disclosure: the same disclosure wording appears in a fixed position at the top of every report.

④ Detailed timeline from the first outflow to the Phase 2 withdrawal reopening

⑤ Bitcoin price paths by incident

Each chart rebases the bitcoin close on the day before the reference date to 100. Past cases are shown from 7 days before to 30 days after the reference date, and Binance from 7 days before to 4 days after the reference date, November 2. All charts use a vertical axis of 60–140 so that drawdowns can be compared across incidents on the same scale. DMM Bitcoin and Bitget are omitted because their loss shares, about 0.02% each, are the smallest.