Bitget's $388M Hack Hit Hot Wallets, Not Cold [2026]

Bitget lost roughly $388 million from its hot and warm wallets on September 24, 2026, and the exchange insists that no private key and no cold wallet was touched. Bitget first told customers that $351.6 million had left its wallets . Further on-chain tracing pushed the final verified figure higher, across 12 wallet addresses. Fortune called it the biggest crypto attack of the year to date. The Bitget hack is also a clean test of what cold storage does and does not protect.

Four weeks on, withdrawals are back, the compensation fund is topped up, and Bitget is shopping for a Thai exchange. The open questions are quieter ones. Which two security appliances were breached, who else runs them, and how did attackers sit inside the network from August 31 without tripping an alarm? Anyone holding coins on a centralized platform should read the answers against their own crypto exchange account security habits.

The loss figure changed three times in a month. The attribution to North Korea remains a suspicion, not a finding. And the single most important detail, the identity of the vulnerable vendor products, is still missing from the public record.

Don't miss new tech stories on Google

Add Shattered once in the Google app and our stories show up in your news suggestions.

What happened at Bitget on September 24

At 18:31 UTC on September 24, 2026, unauthorized transfers began leaving part of Bitget’s hot and warm wallet infrastructure. Funds moved across several blockchains at once. Bitget’s reconciliation system flagged a discrepancy at 19:05 UTC, and the risk-control layer automatically blocked withdrawal requests across the platform. Bitget says deposits and trading kept running while it investigated.

The affected assets included ETH, XRP, BNB, AVAX, USDT and USDC, spread over Ethereum, the XRP Ledger, Arbitrum, Avalanche, Optimism, BNB Smart Chain and Base. Later tracing found additional transactions on Zcash and TRON, which is a large part of why the loss estimate grew. A thief who needs to cash out fast on many networks at once is working from a script, not improvising.

Bitget’s first public statement was brief and, in hindsight, optimistic scope. Bitget CEO Gracy Chen’s initial disclosure read: “At 18:31 UTC on September 24, 2026, Bitget’s security systems identified unauthorized transfers involving a limited number of hot wallets,” as quoted by The Hacker News from the exchange’s post on X. The same post added: “Bitget’s cold wallets and the overwhelming majority of platform assets remain secure and unaffected.”

Both sentences held up on the cold-wallet point. The phrase “limited number” aged poorly once the tally passed $387 million. Bitget also stressed that Bitget Wallet, its separate self-custodial product, runs on independent infrastructure and was not affected.

Bitget hack timeline: the first three hours and the recovery

Bitget’s incident page gives a UTC timeline that is unusually granular for an exchange breach. The detection gap was 34 minutes, from first transfer to the reconciliation alert. The company activated its highest-level emergency response nine minutes after that. The most telling entry is the one at 20:40 UTC. Because private-key compromise had not yet been ruled out, the wallet team began sweeping funds into cold wallets.

Source: Bitget’s incident timeline, last updated October 20, 2026.

Bitget restored withdrawals asset by asset. Bitcoin came back on September 28, Ethereum on September 29 across Ethereum, BSC, Arbitrum, Base and Optimism, and USDT on September 30. The phased plan finished on October 2. Bitget says no further unauthorized transfers have been identified since containment.

How the attackers got in: two security appliances

For the first 48 hours, the entry point was a mystery. On September 30, Bitget published two independent investigations that settled it. BleepingComputer reported that the attackers exploited zero-day flaws in two third-party security appliances, then used that foothold to reach the wallet environment.

Mandiant, Google Cloud’s incident-response arm, found that a threat actor gained privileged access to appliances it labeled A and B on September 24. The actor dropped a web shell on appliance B, set up a command-and-control channel, and moved laterally to Bitget’s production wallet job server, where it deployed malicious packages. SlowMist described a longer dwell time. Its logs show the earliest malicious activity on August 31, when a hidden script on one node read an environment variable holding a database password and connected to the database. Similar activity appeared on two more nodes on September 23 and September 25.

If those logs are right, the intruders had roughly three and a half weeks of quiet access before they moved a single coin. The theft itself ran for nearly three hours, according to SlowMist, and relied on a custom withdrawal tool.

The vendors are still unnamed

The published reports describe the products only as appliances A and B. A Bitget spokesperson did not answer BleepingComputer’s questions the zero-day on the day the reports landed. That leaves every other customer of those appliances guessing whether they run the same vulnerable code. Until a vendor advisory or CVE identifier appears, the practical defense is to treat any internet-facing security appliance as a high-value target and monitor it like a production server.

Related guides: Readers who want something to keep after this story can start with Wallet security in twelve steps , one of our guides in the same area, and the Shattered guides hub groups the rest by task, with the comparisons that explain the trade-offs behind each recommendation.

Spoofed instructions, not stolen private keys

The most useful detail for engineers is how the money moved. During a three-hour livestream on X, Chen said attackers did not steal private keys. They compromised a backend system in the wallet infrastructure, spoofed transaction data, and let Bitget’s own authorization process approve the withdrawals, Fortune reported . In other words, the signing machinery worked exactly as designed. It signed what it was told to sign.

That is a different failure from a leaked seed phrase or a compromised hardware wallet. The cryptography held, and the pipeline feeding it did not. Bitget’s own summary says the attacker used high-privilege internal credentials to send fraudulent withdrawal commands that bypassed existing risk controls.

Three design questions follow for any exchange. Does the signer verify the destination and amount against an independent record before it signs? Do withdrawal limits and velocity checks run on a system the backend cannot rewrite? And does anything alert a human when a hot wallet empties across at least seven chains in under three hours? Bitget’s reconciliation system answered the last one in 34 minutes, which is respectable. It did not prevent the loss.

How much was stolen: $351.6M, $387.5M or $388M?

Readers will see three different loss figures depending on the outlet and the week. They are not contradictory. They are successive estimates as Bitget classified more transactions. The Block carried the $387.5 million revision, which came from additional on-chain tracing, and Bitget’s incident page now states a final verified amount of $388 million involving 12 wallet addresses.

The table exposes one gap. Bitget’s security notice still describes a fund of more than $464 million, while its September 30 post says the fund was restored above a $300 million minimum. Bitget has not published a reconciliation of the two. The likeliest reading is that $300 million is the policy floor and $464 million was the BTC-denominated value of the 5,500 BTC at the time of the notice, but that is inference. Bitget also has not yet published its full-year 2026 report, so the financial hit to the company itself remains unquantified.

Hot, warm and cold wallets: why the three-tier model held

Bitget describes a three-tier custody model. Hot wallets handle live withdrawals, warm wallets refill them, and cold wallets hold the bulk of customer assets offline. Chen’s description of the breach, reported by CoinDesk on September 24, was that the intrusion reached only a portion of the hot and warm layers. The structure is the reason a nine-figure theft did not become a ten-figure one.

For individual holders, the same logic applies at a smaller scale. A balance parked on an exchange is only as safe as that exchange’s weakest operational control. A balance in your own offline wallet depends on different things, namely backups, physical security and firmware hygiene. Our Bitcoin cold storage setup walkthrough covers the offline half of that equation, and the hardware wallet security checklist covers the device half.

What cold storage cannot fix

Cold storage defends keys. It does nothing a compromised instruction path. At Bitget, the attackers never needed the cold keys because the hot and warm layers already held enough value to make the heist worthwhile. The exchange’s decision to sweep funds into cold wallets at 20:40 UTC was a reasonable containment move, since it removed the remaining balance from reach while the team worked out whether keys had leaked.

The uncomfortable corollary is sizing. Hot and warm wallets only need to hold what a normal day of withdrawals requires. Every dollar above that is a standing invitation. Exchanges rarely publish their hot-wallet targets, which is why proof-of-reserves reports, useful as they are, cannot answer the question of how much sat in the exposed layer.

Protection fund and proof of reserves: the compensation math

Bitget established its User Protection Fund in 2022 with 5,500 BTC and a stated $300 million minimum. A 2023 Bitget announcement framed the $300 million level as a way to reassure users after the FTX collapse. On September 24 the company said the incident fell within the fund’s coverage and that customer balances were unaffected. It committed on September 28 to restore the fund to at least $300 million within a week, and announced on September 30 that it had done so, two days early.

Bitget then published its 47th proof-of-reserves update, with a snapshot taken on September 29. It showed an overall reserve ratio of 131% across 19 covered assets, with every covered asset above the 100% benchmark. Reserve ratios above 100% are good news. They do not prove that nothing is missing from the exchange’s operating wallets, and they say nothing liabilities that fall outside the 19 assets.

The fund did its job here. A $388 million loss against a fund of more than $464 million is a covered event, and users saw no haircut. The weak point is the replenishment source. The sources reviewed do not say where the money to rebuild the fund came from, and a fund that depends on the exchange’s own profitability is an insurance policy written by the insured.

The North Korea question: suspicion, not a verdict

Chen said that IP behavior patterns and on-chain analysis made the attack highly consistent with known North Korean hacker organizations. That is an attribution claim from the victim, and Bitget’s official materials do not present it as a confirmed finding from the Mandiant or SlowMist reports. Treat it as a strong suspicion.

Outside analysts have added circumstantial evidence. Elliptic reported links between XRP from the Bitget theft and ETH from an earlier exploit attributed to North Korea, plus connections to addresses used to launder the proceeds of the 2025 Bybit hack. TRM Labs found overlaps with wallets used in earlier North Korean laundering and pointed to the TraderTraitor group. Laundering overlap is useful, but wallet reuse is not proof of identity.

The wider backdrop is real. Chainalysis estimated that North Korean-linked hackers stole a record $2 billion in cryptocurrency in 2025, according to Fortune’s reporting. Our earlier coverage of North Korea’s WaterPlum campaign shows how the same ecosystem also targets individual wallets at scale. The Bitget case fits that pattern of patient, infrastructure-level access followed by a fast cash-out.

Recovery odds look thin. Per a CoinDesk tracking dashboard cited by The Hacker News, Circle and Tether had frozen only $339,100 in stablecoins tied to the hack. Some chain foundations froze attacker addresses. Bitget’s recovery bounty pays 5% of any amount frozen or recovered, which signals how little the company expects to claw back unaided.

Market impact: BGB, the Thai deal and the cost of trust

Bitget’s native token BGB fell almost 7% to $1.93 in the immediate aftermath, then edged back to $1.97, according to Fortune. That is a modest reaction for a nine-figure theft and suggests traders saw the fund and the quick withdrawal restoration as credible. Exchange outflows are the metric to watch, and Bitget has not published them.

The company did not retreat. On October 20 it announced plans to acquire a licensed Thai exchange, reportedly GMO-Z.com Cryptonomics, the operator of Z.com EX, through a Bitget-linked Thai vehicle. Cryptometer framed the move as arriving weeks after the hack. The deal is subject to Thai regulatory approval, and completion was reportedly targeted for October 31. Whether regulators are comfortable approving an acquirer that is still mid-recovery from a breach is a fair question, and the answer will say a lot how Thai authorities weigh incident response.

Bitget is also pushing its Universal Exchange strategy, which adds tokenized equities and other non-crypto markets alongside spot and futures. Wider product surface means wider attack surface. Each new asset class brings more wallet types, more integrations and more third-party software in the path of a withdrawal.

How Bitget compares with other 2025-2026 crypto heists

Placed to recent incidents, Bitget stands out for the layer that failed rather than for the size. The Bybit theft of 2025 reached an ETH cold wallet. Bitget’s cold wallets stayed safe and the hot and warm layers did not. Coldcard, a device built to keep keys offline, lost its reputation for a different reason, as our hardware wallet security analysis after the Coldcard hack explains.

Sources: Fortune for Liquid Network and Coldcard, BleepingComputer for Bybit, Bitget for its own figures. Loss figures are as reported by those outlets at publication.

Each row hits a different custody model, and none of them is immune. Exchange custody concentrates risk but offers a compensation fund. Hardware wallets remove the operator but put the burden on firmware and the user. Multisig setups spread signing power but can still approve a bad payload. Our guide to a Safe multisig setup after the Bybit hack walks through that last category.

Historical context: from FTX to Bybit to Bitget

Exchange failures have shaped the industry’s safety nets for a decade. After FTX collapsed in 2022, exchanges rushed to publish reserve data and set up user funds. Proof-of-reserves reporting spread across the sector in the same period, and Bitget’s updates now use Merkle tree verification so users can check their own balances against a published root.

Those measures target insolvency and mismanagement. They are weaker against a skilled intruder inside the network. The Bybit heist of 2025 showed that a state-linked group could take $1.5 billion from a cold wallet. Bitget shows a second route, which is to live inside the operations stack for weeks and then replay the exchange’s own approval process. Reserve proofs would have looked fine on September 23.

Each cycle has pushed exchanges toward more segregation. After this one, the pressure will land on vendor risk. A security appliance sits at the top of the trust chain, with privileged network access and often its own update channel. Compromising two of them gave the attackers a path that no wallet-level control was designed to see.

What users and builders should do now

For ordinary users, the Bitget outcome was benign because a funded backstop existed. Do not assume every exchange has one. Check whether a platform publishes a protection fund, how large it is relative to customer assets, and how recently it updated its proof of reserves. Keep only trading balances on an exchange, and move long-term holdings to wallets you control. Our crypto wallet security steps cover the migration, and a tool for tracking crypto wallet alerts will tell you when a watched address moves.

For exchange engineers and security teams, the Mandiant and SlowMist findings suggest a short list.

Inventory every third-party security appliance with network access to wallet systems, and log who patches each one.

Hunt for web shells and hidden scripts on those devices, and treat environment variables and database credentials as potentially exposed.

Verify withdrawal instructions out of band, so a compromised job server cannot approve its own transfers.

Rotate credentials after any appliance advisory, even if no direct intrusion is evident.

Alert on hot-wallet drain patterns across chains, not just per-chain thresholds.

Bitget’s own recovery steps are a decent template for the incident-response side. It isolated systems, moved remaining funds to cold storage, brought in two independent investigators within a day, and published a timeline that outsiders could check.

Five predictions for exchange security

These are analysis, not reporting. They follow from the facts above but carry no official confirmation.

Vendor disclosure arrives within weeks. Two zero-days in security appliances will not stay anonymous. Expect an advisory or CVE from the affected vendors, and expect other crypto firms to audit their own deployments first.

Independent transaction verification becomes a baseline. Large exchanges will add out-of-band checks that confirm destination and amount before any signer acts, copying what Bitget’s incident exposed as missing.

Proof-of-reserves gets a companion. Auditors and exchanges will push for reporting on hot-wallet sizing and access controls, because reserve ratios alone did not warn anyone here.

The Thai deal faces tougher questions. Regulators will likely ask for incident-response evidence before approving the October 31 target, so a slip past that date would not be surprising.

Attribution firms converge on a name. With Elliptic and TRM Labs already pointing at DPRK-linked laundering routes, a formal government attribution is plausible before year-end. Our wider cryptocurrency coverage will track it.

Frequently asked questions the Bitget hack

How much did Bitget lose in the hack?

Bitget’s final verified figure is $388 million, involving 12 hot or warm wallet addresses. The first estimate on September 24 was $351.6 million, and a $387.5 million revision followed after more on-chain tracing.

Were customer funds lost?

Bitget says user balances were unaffected and that its User Protection Fund covers the loss. It completed phased withdrawal restoration on October 2, 2026.

Were private keys or cold wallets compromised?

Bitget says no. Its notice, updated October 23, says cold wallets remained secure and the breach was confined to part of the hot and warm layers. Attackers spoofed transaction data and triggered Bitget’s own authorization process.

How did the attackers get in?

Mandiant and SlowMist reported zero-day exploits against two third-party security appliances, followed by lateral movement to a production wallet job server. SlowMist’s logs show the earliest malicious activity on August 31. The vendors have not been named publicly.

Was North Korea behind it?

Bitget’s CEO said the attack pattern was highly consistent with North Korean groups, and Elliptic and TRM Labs reported laundering overlaps. Bitget’s official materials do not treat the attribution as a confirmed finding, so it remains a suspicion.

Is Bitget safe to use now?

Withdrawals work, the fund was replenished above $300 million on September 30, and a September 29 proof-of-reserves snapshot showed a 131% ratio. Safety is still a judgment call. Keep only the balances you need for trading on any exchange.

Can the stolen funds be recovered?

Some chain foundations froze attacker addresses, and Circle and Tether froze $339,100 in stablecoins, per a CoinDesk dashboard cited by The Hacker News. Bitget offers a 5% bounty for funds frozen or recovered. Large-scale recovery looks unlikely so far.