Bitget $352m Hack: Why Exchange Security Is Guarding the Wrong Layer
Attackers took $351.6 million from Bitget without touching a private key, by forging the instructions its signing machines trusted. Bybit lost $1.5 billion the same way 19 months ago. The industry spent years protecting keys, and the losses have moved to the systems that tell those keys what to sign.
The most reassuring line in Bitget's account of its hack is also the most worrying. "Private key compromise has been ruled out," chief executive Gracy Chen said on Friday, after attackers moved $351.6 million out of the exchange's hot and warm wallets on Thursday evening. The keys never leaked. The attackers compromised a third-party tool that Bitget used every day, forged transfer instructions and got Bitget's own signing machines to approve them.
This is the second time in two years that one of the largest exchanges has been emptied by feeding its signers false information. Bybit lost $1.5 billion in February 2025 after North Korea's Lazarus Group compromised the Safe multi-signature interface its staff used, leading staff to approve a transaction they could not see. Bitget's version is more automated. The attackers talked to machines. The weak point is the layer between "what we think we are signing" and "what we are signing".
The industry's standard security pitch now aims at the wrong target. Cold storage, multi-signature schemes, hardware security modules and monthly proof-of-reserves reports all answer one question: can someone steal the keys? The answer has improved a great deal. The losses that matter now come from another question: can someone persuade the keys to sign? Exchange marketing does not address it, and no reserve attestation can measure it.
Bitget's systems flagged unauthorised transfers at 18:31 UTC on September 24. Over roughly an hour, funds in ether, USDT, USDC, AVAX, BNB and XRP left wallets across seven networks, including Ethereum, Arbitrum, BNB Smart Chain and Base. Roughly $183 million came off EVM chains and about $143 million in XRP, on-chain trackers' tallies show. Within six minutes, one fresh address converted 19.67 million USDT into 7,111 ETH on Arbitrum. The attacker paid up to 5% above market through UniswapX and 1inch Fusion to do it. That premium was cheap insurance, because Tether and Circle can freeze their stablecoins at the contract level and nobody can freeze ether. CoinDesk has Chen's account of the mechanism.
Chen says North Korea is "very likely" responsible, citing IP addresses tied to VPN services a known DPRK group has used. The swap-to-ether playbook matches Bybit's. If the attribution holds, Pyongyang has now taken most of the year's stolen crypto from three targets. TRM Labs put first-half 2026 hack losses at $972 million, with $643 million attributed to North Korea, mostly the $292 million KelpDAO bridge attack and the $285 million Drift breach. TRM's tally shows a state actor running a small number of large, well-researched operations. Bitget would lift the attributed total for the year to roughly $1 billion.
Deposits and trading are open. Withdrawals are suspended, and Chen has declined to give a date for reopening. BGB, the exchange's token, fell about 3% to $1.97. The market is taking the loss calmly.
That calm rests on Bitget's protection fund, which the company says holds 5,500 bitcoin worth about $464 million and covers the full loss. It does this time. Paying out $351.6 million would use about three-quarters of the fund, leaving close to $112 million until the exchange refills it. The fund is held in bitcoin, so its capacity shrinks when the market does, which is also when exchanges tend to be most stretched.
A protection fund is self-insurance: the exchange sets aside its own balance-sheet assets and chooses what they are held in. It is better than nothing, and it is why Bitget customers are not facing a haircut today. It also explains why the exchange's August proof-of-reserves report, its 45th monthly disclosure, which showed a 122% reserve ratio, told users nothing about this risk. Bitget's August report proves the assets were there on the snapshot date. It says nothing about whether a compromised vendor could instruct the signers to send them elsewhere a month later.
The defence of Bitget is that the system contained the breach. Hot and warm wallets were affected, cold storage was not touched, customers will be made whole and the exchange kept trading. Bybit made the same argument in 2025, reopening withdrawals within about 72 hours with bridge loans, and it is still one of the largest venues in the market. On that reading, a well-capitalised exchange can absorb even a nine-figure theft as an operating cost.
An exchange that can lose a third of a billion dollars in an hour to a supplier compromise has a control problem. A large balance sheet only lets it keep having that problem. If authorisation trusts whatever the upstream system tells it, the size of a hot wallet sets the maximum loss, not the quality of the security. The fix is known: signers that check each transaction against an independent record of what was requested, per-destination rate limits and time delays on large transfers to new addresses. Each one makes withdrawals slower. That trade-off is why exchanges have resisted, and why the losses keep arriving.
A smaller story on Thursday makes the same point at the other end of the crypto plumbing. The US has seized the bank accounts of a payments group linked to Tether, the Financial Times reported. Prosecutors are seeking to forfeit about $84.2 million held at Wells Fargo, JPMorgan Chase and two crypto addresses, tied to Capstone, a payments firm they allege misrepresented its business to US banks while moving money for EQIBank, a Dominica-licensed digital bank. EQIBank processed wires for USDT purchases and redemptions. It says the seizure has frozen about 80% of its cash and warned on September 9 that it could be forced into liquidation. The Crypto Times has the court detail.
Tether puts its trapped funds at under 0.034% of group assets, or at most about $64 million out of $187.75 billion, and says it knew nothing of Capstone's conduct. The number is trivial and USDT's peg isn't at risk. The route is what matters. The world's largest stablecoin relied on a small Caribbean bank that relied on a US processor now accused of misleading its own banks. The token worked as designed, and the off-chain chain of trust around it failed.
Bitget has to reopen withdrawals and publish a technical account of which vendor was compromised and how its signers were fooled. We expect withdrawals to resume before the end of next week. The post-mortem will be the more telling document. Unless it names hard limits on what the signing machines will accept, independent of the systems that request transfers, the next nine-figure loss will come through the same door. It is likely to come before the end of this year.