Bitget is about to reopen withdrawals: after the theft of 380 million USD, the real test is just beginning

Author: Gu Yu, ChainCatcher

On September 26 at noon, Bitget fulfilled its promise from the previous day and announced a phased withdrawal recovery schedule: withdrawals for Bitcoin will open on September 28 at 16:00 (UTC+8), Ethereum, BSC, Arbitrum, Base, and Optimism ETH will open on September 29 at 16:00, USDT on Ethereum, BSC, Solana, and Tron will open on September 30 at 16:00, and the remaining tokens, fiat, and P2P will open on October 2 at 16:00. The official statement also reiterated that the suspension of withdrawals "is a security measure and is unrelated to the availability of user assets," and that deposits and trading have never been interrupted, with financial losses covered by the protection fund.

From the detection of the first unauthorized transfer at 18:31 (UTC) on September 24 to the opening of the first batch of withdrawals, about 86 hours elapsed; by October 2, everything was restored, which took seven and a half days.

This duration has a reference point. In February 2025, Bybit was hacked for $1.5 billion, four times the scale of this incident, and Ben Zhou filled the gap with bridge loans, restoring withdrawals within 72 hours. Bitget's incident was smaller in scale but took longer. Bitget's Chinese region head, Xie Jiayin, explained, "Due to the different methods of theft, to absolutely eliminate all potential risks, the handling method is also different from other cases."

This statement actually reveals something more important: the issue is not with the vault itself, but with the entire process. Fixing the process is much slower than changing a lock.

From 180 million to 387.5 million: The numbers are still being revised, but it took nearly three hours to stop the bleeding

The amount has been revised twice so far and is likely to change again.

The initial report was $351.6 million, and previous estimates from on-chain institutions were even lower—Arkham analyst Emmett Gallic initially estimated $178 million, while PeckShield and Bubblemaps estimated between $145 million and $192 million, which were only about half of the final figure. On September 25, Bitget further revised the amount to approximately $387.5 million, citing "a more complete accounting of transfer situations during the incident," adding Zcash and TRON on-chain amounts of ZEC, TRX, and XAUt that were previously not counted. The official emphasized that this does not indicate new thefts, and the incident is under control.

According to Lookonchain's breakdown, the bulk of the funds is unusually concentrated: 102,930,000 XRP (approximately $157.48 million), 31,890 ETH (approximately $85.75 million), 34.75 million USDT, 21.05 million USDC, 19.67 million USDT0, 3,000 XAUt (approximately $12.82 million), 12,719 BNB, 821,012 AVAX, and 20.59 million TRX. The marked list from SlowMist's MistTrack shows that the attacker's address includes at least seven Ripple addresses (totaling approximately 102,926,478 XRP), 11 EVM addresses, and one TRON address.

What is truly worth investigating is the timing. On-chain data detected the problem earlier than the official announcement: before 19:00, Arkham analyst Emmett Gallic, PeckShield, and Bubblemaps had already marked the anomalies. Arkham later calculated that $228 million left the Bitget wallet within 18 minutes from 18:58 to 19:16. Meanwhile, on-chain data cited by Forbes shows that the outflows continued for nearly three hours, with the last ETH transfer occurring 52 minutes before the official announcement.

In other words, it took nearly three hours from "detection" to "stopping the bleeding." During these three hours, the attacker converted most of the loot from the EVM side into approximately 67,982 ETH (about $183 million)—not for investment, but to make themselves "unfreezable."

There is also a contradiction that has not been resolved: Arkham claimed that $153 million of XRP came from an address it labeled as Bitget's cold wallet, while Bitget has consistently insisted that the cold wallet is completely safe. As of now, the statements from both parties have not aligned publicly.

The private key was not lost: What was bypassed was the authorization process

Gracy Chen's explanation given in the early hours of September 25 is the most unusual part of this incident.

She stated that the attacker compromised the core backend system of the wallet service, thereby "generating false transfer information and invoking Bitget's signing process," allowing funds to be transferred out of the platform; she also clarified that "the possibility of private key leakage can be ruled out." During a three-hour live stream on X that day, she further mentioned that preliminary clues point to a supply chain attack—what was compromised were the third-party tools used daily by the exchange, rather than its core system, and there is currently no suspicion of internal personnel involvement. Mandiant and SlowMist are still assisting in the investigation, and the specific intrusion path awaits confirmation in the final report.

The significance of this technical path lies in its redefinition of security boundaries. For many years, the narrative around exchange security has revolved almost entirely around private key management: hot and cold separation, multi-signature, MPC, HSM. Bitget itself has used this language—three-layer wallet architecture, hot, warm, and cold separation, with cold wallets offline. But this time, not a single private key was lost, and not a single line of code in the cold wallet was touched; the attacker simply made the system process a forged withdrawal request as a normal request without anyone noticing.

Thus, the question shifted from "where is the money stored" to "who has the authority to move the money."

The industry's reaction has been quite restrained, but not mild. Esme Pau, head of capital markets and policy at CertiK, stated that this is "one of the most substantial centralized exchange attacks of 2026," a "wake-up call for the digital asset industry," and pointed out that the scale of the theft is equivalent to three-quarters of the protection fund, "surpassing ordinary security lapses, turning it into a crisis event." Aneirin Flynn, CEO of cybersecurity company FailSafe, put it more directly: the significance of this theft lies in "destroying the illusion that large exchanges have solved hot wallet security"; even if the protection fund covers the losses, "such a scale of intrusion still severely damages institutional trust in crypto infrastructure."

Ledger CTO Charles Guillemet pointed the finger at the variables themselves—AI is making it cheaper to discover vulnerabilities and write attacks, "security has always been a challenge in the crypto space, and AI is shifting the advantage to attackers. Closing this gap requires not just patching up the aftermath of the next attack, but changing the underlying principles and architecture of security."

Looking at these three points together, the attack surface for 2026 is already clear: CertiK has observed that the focus of attacks is shifting from smart contract vulnerabilities to infrastructure intrusions, hot wallet thefts, and key leaks. The difference is that contract vulnerabilities can be exhaustively audited, while the trusted data chain in the approval process cannot be exhaustively enumerated in advance to determine which link will be forged.

0.08%: The efficiency and ceiling of recovering stolen funds

How much of the stolen funds can be recovered was almost decided within the first few minutes of the incident.

Circle and Tether both took action. On September 25 at 05:00 (UTC), Circle invoked the built-in blacklist feature of the USDC contract, freezing 99,990 USDC in the address marked by Etherscan as "Bitget Exploiter 8"; about seven hours later, Tether added the same address to the USDT blacklist, locking 218,023 USDT. In total, about $318,000, accounting for approximately 0.08% of the total stolen amount.

This percentage is glaringly low, but it does not reflect negligence on the part of Circle and Tether; rather, it illustrates the structural ceiling of the situation: stablecoins are contracts written by the issuer, and the issuer can freeze at the contract level; ETH is the native asset of the network, with no contract, no issuer, and no one has that authority. The attacker clearly understood this better than anyone else—within six minutes, a newly created wallet converted 19.67 million USDT0 into 7,111 ETH through UniswapX and 1inch Fusion, paying a premium of about 5%. To remain unfreezable, they were willing to sell at a loss. On-chain analysis disclosed by Bitget also shows that within hours of the incident, approximately 6,300 ETH (nearly $19 million) flowed into Tornado Cash.

The consensus among trackers is that the attacker’s associated addresses still hold over 63,000 ETH, beyond the reach of any issuer.

What remains is a recovering network that is taking shape but has limited efficiency. Bitget launched a recovery bounty program, offering a 5% reward to those who assist in freezing or recovering funds, while also launching a real-time tracking dashboard, information submission portal, and attacker address API, and supporting the submission of leads through Bybit's Lazarus Bounty platform. In the industry, Bybit CEO Ben Zhou stated he is ready to help—Bitget had assisted in the investigation after his $1.5 billion theft; Binance co-CEO Richard Teng said they have shared intelligence and assisted in tracking; MEXC CEO Vugar Usi expressed readiness to support Bitget in any way; and CZ also publicly voiced support.

This scene of "industry peers helping each other after a theft" has formed rapidly over the past two years. However, what it can do is always limited to freezing the portion of funds that could not be converted to ETH in time.

How much is left in the protection fund

A more practical question is: is the $464 million protection fund sufficient?

Based on the revised $387.5 million, the coverage ratio is about 1.2 times, with only about $76.5 million remaining as a buffer—more than a third less than the initial estimate of $351.6 million. Gracy Chen attempted to reinforce this commitment with more numbers: "In addition to this protection fund of over $464 million, all stored in publicly verifiable wallets, Bitget itself also holds over $1 billion in assets, with user funds covered at a 1:1 ratio." Bitget has released reserve proofs for 45 consecutive months, with an August reserve ratio of 122%.

All of this falls into the category of "company statements," and everyone knows what this means after FTX. The real testing points are two: whether a complete event report can be released on time, and whether users can actually withdraw their funds after September 28. History has also provided negative examples—when Bitmart was hacked for $196 million in 2021, it similarly promised full compensation, but by 2026, the platform had shut down, and users still had not recovered their funds.

In terms of attribution, Bitget claims that the IP behavior patterns and on-chain characteristics point to North Korea. Elliptic determines this to be the largest suspected North Korean cryptocurrency theft case of 2026, pushing the total amount of stolen cryptocurrency assets for the year to over $1 billion. Chainalysis data shows that North Korea had already stolen a record $2 billion in 2025 and is increasingly relying on dispatching IT personnel to infiltrate target companies. SentinelOne had just linked the TraderTraitor------Bybit $1.5 billion case and the KelpDAO $292 million case to the same group behind an intrusion targeting an Indian IT service provider a week before this incident. However, some analysts remain cautious about directly linking it to Lazarus, viewing it as an investigative lead rather than a law enforcement conclusion.

Currently, the Bitget platform token BGB has fallen from around $2.02---$2.06 before the theft to about $1.93, a decline of approximately 5%---7%, before slightly rebounding; during the same period, the overall cryptocurrency market has actually risen nearly 10% over the past week.

Conclusion

In the past two years, the cryptocurrency industry has made significant efforts to convince the outside world that "we are different now"------licenses, audits, proof of reserves, protection funds, three-tier wallet architecture, Bitget has all of these, recently announcing a 122% reserve ratio, 45 months of uninterrupted Proof of Reserves, and even a protection fund that is still sufficient for payouts.

However, the path to losing money this time does not lie in any of these items on the list. The attackers did not crack any cryptography or steal any private keys; they simply allowed a transfer request that should have been reviewed to pass through an entire approval process without obstruction to the final step. What the industry has fortified is the keys, but what has been breached is trust; and trust can never be cold-stored.

Therefore, the gate that opened on September 28 has significance far beyond just the restoration of withdrawals. It is also a stress test: if users can withdraw their money in full within the promised timeframe, Bitget will turn this crisis into a Bybit-style trust rebuilding; if there are queues, limits, or any form of delays in withdrawals, then the timetable that took 86 hours to establish will ultimately measure not the speed of technical fixes, but the ability of an exchange to fulfill its commitments.

After a theft, the cause of death for an exchange is never because it lost money, but because users find they cannot withdraw their funds.