Every time a large exchange freezes withdrawals, the same question follows within the hour. It followed Bitget this week, after it disclosed unauthorized transfers of $351.6 million and paused withdrawals while it investigated.
The question is worth answering carefully rather than dismissing, because the two failure modes it conflates are entirely different. FTX did not collapse because someone broke in. It collapsed because the money customers thought was theirs had already been spent. Bitget’s position is that the money is still there, that the shortfall is smaller than a fund set aside for exactly this, and that cold storage was untouched. Those are claims that can be tested, and the next few weeks will test them.
What FTX Actually Was
FTX failed as a solvency and integrity collapse, not a security one.
Customer deposits were treated as available capital. Alameda Research, the affiliated trading firm, used customer assets and the exchange’s own FTT token to cover losses and support leverage in an arrangement where the collateral was a token the exchange itself issued. When reporting on Alameda’s balance sheet circulated in early November 2022, confidence went. Withdrawal requests surged, the platform could not meet them, and bankruptcy followed within days. The criminal case that followed established fraud rather than mismanagement.
Three absences mattered. FTX published no durable, user-verifiable proof-of-reserves program. It had no separately disclosed protection fund larger than its shortfall. And when the run began, there was nothing anyone could point to in real time.
What Bitget Says It Has
According to its security notice, the incident involved a compromised backend system that spoofed transfer data and triggered the exchange’s own authorization process, draining hot and warm wallets while cold storage was unaffected. Chief executive Gracy Chen said in a post on X that private keys were not compromised. She later said the company was working with Mandiant and SlowMist alongside law enforcement.
The distinction between wallet tiers matters. Hot wallets are the operational cash drawer, kept online so withdrawals can be processed quickly. An attacker who can inject false instructions into the signing pipeline can empty that drawer without ever holding the master keys. That is a serious failure of internal systems. It is not the same as the assets not existing. Within the loss, XRP made up the largest single share.
Bitget launched monthly Merkle-tree proof of reserves in December 2022, weeks after FTX’s collapse. Its September 2026 report was the 46th consecutive monthly disclosure, showing a total reserve ratio of 135% across 19 assets. Users can check whether their own balance was included using an open-source Merkle validator.
Bitget is explicit about the limits of that mechanism. Its own documentation states that a reserve ratio above 100% should not be treated as the equivalent of a financial audit or a guarantee of future solvency. A Merkle proof shows a user’s balance was included in a snapshot of claimed liabilities and that on-chain reserves exceeded those liabilities at that moment. It does not audit off-chain arrangements, related entities, or obligations that never entered the snapshot.
The Protection Fund is the other pillar, and it is a separate disclosure from proof of reserves. Bitget established it in 2022 with a $300 million commitment. It is backed by 5,500 BTC, so its dollar value moves with Bitcoin’s price: in August 2026 it averaged $382 million, ranging from about $345.3 million on August 1 to $441.5 million on August 27.
Where the Comparison Breaks Down
FTX could not pay because the money was gone. Bitget’s public position is that the money is on its books, the shortfall is ring-fenced by a disclosed fund exceeding it, and the majority of assets were never touched.
If that is accurate, the situations are not analogous. A security breach covered by a reserve is an operational failure with a financial buffer. An $8 billion hole with no buffer is insolvency.
It is also true that Bitget’s leadership is more visible than FTX’s was and that its reserve reporting has run for nearly four years with no equivalent at FTX. Chief executive Gracy Chen said this week that she does not believe major exchanges today would intentionally commingle customer funds and that user assets are backed one-to-one and verifiable through proof-of-reserves data.
Where It Doesn’t
A withdrawal freeze looks identical from the outside whatever causes it. Users who lived through November 2022 cannot verify, in the moment, whether a pause is forensic caution or the start of something worse. That anxiety is rational, and telling people it is misplaced does not make the freeze shorter.
The published safeguards are also promises until they are used. A protection fund matters only if the assets in it are liquid and unencumbered, and that has not been tested at this scale. A reserve ratio matters only if it still holds after the loss is fully accounted for.
Some risks here are ordinary rather than exceptional. Hot-wallet architecture and third-party tooling create attack surface that self-custody does not. Expansion into tokenized equities and other instruments adds operational complexity. And regional licensing is uneven: Bitget announced in August that it would wind down services for Japanese residents, one example of how recourse varies by jurisdiction. None of that is evidence of fraud. It is the standing risk profile of large centralized venues.
What Would Settle It
The honest answer today is that the situations are structurally different and that the difference depends on claims not yet verified independently.
Three things would confirm Bitget’s account: withdrawals reopening promptly and processing normally; the Protection Fund being visibly drawn down, with the movement traceable on-chain; and the next proof-of-reserves report still showing a surplus once the incident is reflected.
Three things would undermine it: withdrawal delays extending without clear explanation; communication becoming vaguer rather than more specific; or reserve ratios changing materially after reconciliation.
Independent verification may be thinner than usual. Security researcher ZachXBT said he would not monitor this incident, removing one of the outside analysts users have come to rely on in cases like this.
Those are observable over weeks, not hours. Until then, what can be said is narrow: the loss is disclosed, the fund is claimed to exceed it, cold storage is said to be intact, and none of that has yet been through the one test that matters, which is a large number of users asking for their money at once.
FTX’s lesson was that confidence can disappear faster than facts can be established. The corollary is that facts, when they arrive, are more useful than either reassurance or alarm.
