Bitget Lost $387 Million Without a Single Private Key Being Stolen

Corrected 2026-10-15: An earlier version of this post said we build tokenized-asset infrastructure. Our tokenization work was an R&D demo (Acre Estate), parked since 2026-10-07. The argument is unchanged.

We’ve designed tokenized-asset systems in our own R&D, so when an exchange loses $387.5 million without anyone actually stealing a private key, that’s not background crypto-Twitter noise — it’s a direct data point on the exact failure mode any wallet architecture we design has to defend against. Bitget was hit by a breach detected September 24, 2026 at 18:31 UTC, and within a day the confirmed loss reached $387.5 million, making it the largest crypto hack of 2026, per CoinDesk and Decrypt.

The number climbed fast, then the story got more interesting

Initial internal detection: $183 million within the first hour.
Public announcement figure: $351.6 million.
Final confirmed tally by September 25: $387.5 million.
The single largest component: roughly 103 million XRP, worth about $157 million.

That escalation pattern — three different loss figures in under 24 hours — is itself a tell that this wasn’t a clean single-transaction theft. It was an investigation racing to find out how much of the wallet infrastructure had actually been touched.

How it actually happened

This is the part that should worry anyone running custody infrastructure more than the dollar figure does. Bitget CEO Gracy Chen described attackers who “compromised a critical backend system within our wallet infrastructure, used it to spoof transaction data, and triggered our authorization process to move funds out,” per CoinDesk and Cybersecurity News. That’s not a key theft — it’s closer to forging a bank’s internal withdrawal request than stealing a vault key. The unauthorized transfers came from hot and warm wallets; Bitget’s offline cold storage was untouched.

The reason that distinction matters: most wallet-security architecture, tokenized-asset custody included, is built around the assumption that the private key is the attack surface — tier the keys, air-gap the cold storage, and you’ve contained the risk. This attack didn’t go after the keys at all. It went after the system that decides whether a transaction request is legitimate. A well-tiered wallet architecture can still be defeated if the backend that approves transactions is compromised, which means “our keys are safe” is not the same claim as “our system can’t be robbed.”

Attribution: North Korea, not yet formally confirmed, but the circumstantial case is strong

Chen said attackers used “IP addresses that match the VPN choices by a certain DPRK group,” and that the operational pattern “looks very much like what the North Korean team did before,” per CoinDesk.
On-chain researchers flagged a freshly created wallet that spent $19.67 million in stablecoins to buy 7,111 ETH in six minutes, roughly 5% above market price via DEXs — a rapid-conversion laundering pattern consistent with prior North Korean operations, per Fortune.
North Korea’s Lazarus Group (tracked under the alias TraderTraitor) is already responsible for the industry’s largest-ever heist, the $1.5 billion Bybit hack in February 2025, which the FBI formally attributed to North Korea, per sanctions.io.
Bloomberg reports this attack pushed North Korea’s cumulative 2026 crypto theft past $1 billion, per Bloomberg.

The bigger pattern this fits into

North Korean-linked hackers were reported to have stolen $643 million in crypto in just the first half of 2026 — worth noting that figure comes from Bitget’s own prior threat reporting, so treat it as a company-sourced estimate rather than independently audited, per Bitget News.
One tracking estimate put North Korean hackers behind 76% of all crypto hack value through April 2026, per sanctions.io.
Other 2026 incidents attributed to the same actors include a $292 million Kelp DAO exploit in April.
The laundering pipeline is now well-documented: rapid cross-chain movement through intermediary wallets within hours, mixers and privacy coins outside US jurisdiction, chain-hopping across Ethereum, Avalanche, BSC, and Bitcoin via unregulated bridges, then OTC conversion to fiat through desks in Southeast Asia and the Middle East with limited AML coverage.
OFAC’s SDN list, UN Panel of Experts reporting, aligned EU/UK sanctions, and FBI wallet blocklists are all in place — none of it has meaningfully slowed the pace of these thefts.

Bitget’s response

Bitget’s User Protection Fund, holding over $464 million, is covering the full loss, and the company says customer balances are unaffected. Deposits and trading continued uninterrupted; only withdrawals were frozen as a precaution, with a September 26 resumption target. Bitget has committed to publishing a full incident report with root-cause analysis once remediation is complete, per Decrypt.

The tension we’re not going to paper over

The comfortable read of this story is “cold storage worked, so the defense-in-depth model is validated.” We don’t think that holds up. Cold storage protected the assets that were never in the path of the attack — it says nothing about whether the approval logic guarding the hot and warm layers is sound, and here it demonstrably wasn’t. Whether Bitget’s full incident report reveals a supply-chain or insider-access vector (rather than a pure external exploit) will determine how alarming this actually is industry-wide: an insider-adjacent compromise of authorization infrastructure is a much harder problem than “someone found a bug.” And recovery rates on DPRK-linked thefts have historically been low, so the fact that the loss is covered by Bitget’s reserve fund doesn’t mean the stolen $387.5 million comes back — it means Bitget, not North Korea, is the one absorbing it.

What we’re watching

Whether Bitget’s incident report names a supply-chain or insider-access vector.
Whether any of the stolen $387.5 million gets frozen or recovered.
Whether this accelerates industry adoption of authorization-layer hardening — multi-party computation, hardware-enforced approval flows — the kind of control that could have stopped a spoofed-authorization attack even with the backend compromised.

For us, the design takeaway is narrow and specific: key custody and transaction authorization are two separate security problems, and an architecture that only hardens the first one is exactly as exposed as Bitget’s was. That’s not a hypothetical for anything we design in this space — it’s the actual threat model.