Mt.Gox: The Forensic Timeline of 850,000 Lost Bitcoin, 2010–2024

Prologue: A name that should never have been used

The name Mt.Gox first appeared on the internet in 2007. Jed McCaleb had written a small website where players of Magic: The Gathering could list their digital cards for trade. The full name was "Magic: The Gathering Online Exchange," abbreviated to MTGOX. At that point bitcoin did not yet exist as a public network, and the phrase "crypto exchange" was not part of anyone's vocabulary.

The name never changed. By the time Mt.Gox was the world's largest bitcoin exchange, handling more than 70% of all bitcoin transactions, it still carried that acronym.

In February 2014 the company announced that roughly 850,000 BTC were missing. Creditors then spent a decade in legal proceedings, and base repayments began in July 2024; the BTC that creditors got back by quantity was worth more, at 2024 prices, than what they had deposited.

This volume goes in chronological order: where the exchange came from, how it took most of the market, where the coins started going missing, and how it finally went down. Three things get their own treatment:

• The June 2011 breach that crashed BTC from $17 to one cent on Mt.Gox's order book, and the theft of the hot-wallet keys that September (Chapters I and III).
• Transaction malleability — the explanation Mt.Gox itself gave, and how much it can actually account for (Chapter IV).
• The Wizsec / Kim Nilsson on-chain investigation that traced the stolen funds to BTC-e (Chapter VII).

Chapter VI sets BitGrail's 2018 collapse alongside Mt.Gox; Chapter X covers where the distributions stand and what is still unresolved.

TL;DR · Key Numbers

If you are short on time, jump to Chapter XI: What to take away from Mt.Gox.

Chapter I: From card-trading site to bitcoin exchange (2007-2011)

From card site to bitcoin exchange (2007–2010)

Jed McCaleb in 2007 was already a serial entrepreneur: he had built eDonkey2000, an early peer-to-peer file-sharing network, and Overnet. MTGOX was a site he built for trading Magic: The Gathering cards; it never took off, and he stopped running it.

In July 2010, McCaleb read about bitcoin on Slashdot and repurposed the old MTGOX domain as a bitcoin exchange; Mt.Gox launched as one on July 18. The name was not chosen for crypto at all — the domain simply happened to still be in his hands.

Karpelès takes over (2011.03)

In March 2011, McCaleb sold Mt.Gox to a French programmer who lived in Tokyo: Mark Karpelès. McCaleb himself went on to co-found Ripple and then Stellar. Karpelès ran Mt.Gox as a Japanese kabushiki kaisha with offices in Shibuya, Tokyo.

June 19, 2011 — the first major breach

On June 19, 2011, an attacker used credentials taken from a Mt.Gox auditor's compromised computer, placed a huge sell order, and drove the price of BTC on Mt.Gox from $17 down to $0.01. Mt.Gox's response was a rollback: it reversed the trades that had executed after the price collapse.

Something with larger consequences followed that September. The Wizsec 2017 report states that in September 2011 the private keys to Mt.Gox's hot wallet were stolen by way of a copied wallet.dat file (more on this in Chapter III). The report also spells out how the books went on skewing. Because addresses were reused, Mt.Gox's system read part of what the thief spent as incoming deposits, credited large BTC amounts to several user accounts on the back of that, and added roughly 40,000 BTC to the imbalance in the process.

Chapter II: How Mt.Gox swallowed 70% of global volume (2011-2013)

Mt.Gox's dominance between 2011 and 2013 rested on three reinforcing factors:

1. First-mover advantage. Mt.Gox launched in July 2010, an entire year before BTC-e or Bitstamp. In the small early-bitcoin community, users tended to identify accounts by reputation, not by platform credentials. Through early 2011, Mt.Gox was effectively the only named venue. Other exchanges existed but had no comparable user trust.
2. US dollar fiat on-ramp. Mt.Gox let users fund accounts in US dollars; the US payment service Dwolla was one of the channels it used.
3. Price index status. Throughout 2012-2013 most financial-media reporting on "the bitcoin price" was, by default, citing Mt.Gox. This was self-reinforcing: deeper books at Mt.Gox produced more reliable price formation, which attracted more reference citations, which attracted more flow.

These three factors compounded. First-mover advantage brought in market makers and large traders. Market makers brought liquidity. Liquidity produced "the canonical price." Financial media cited that price. New users joined Mt.Gox because the canonical-price venue felt safer. Through 2011-2013 the loop built a near-monopoly.

On April 10, 2013, BTC printed $266 on Mt.Gox. Mt.Gox then suspended trading from April 11 to 12 for what it called a market "cooldown."

In May 2013 the US Department of Homeland Security seized Mt.Gox's Dwolla account (see the next chapter); on June 20, Mt.Gox suspended USD withdrawals.

Complaints about delayed withdrawals grew from then on. This was the earliest sign that could be observed from outside.

Chapter III: Where the cracks began (2011-2013)

The question of when Mt.Gox actually began losing customer funds remains formally contested. Wizsec's investigation concluded that the hot-wallet theft began in 2011 and ran until mid-2013. Karpelès, at trial, maintained that the bulk of the missing BTC was taken without his knowledge.

The hot wallet copied out in September 2011

Wizsec's 2017 write-up states it plainly: in September 2011, the private keys to Mt.Gox's hot wallet were stolen, in the form of a simple copied wallet.dat file. From then on the thief could spend the coins those keys controlled; by mid-2013, when the funds spendable from the compromised keys had slowed to a near halt, the thief had taken out about 630,000 BTC.

In other words, over those two years the gap between Mt.Gox's books and its real on-chain balance kept widening; when the company became aware of it has never been established.

The 2013 US DHS seizure

In May 2013 the US Department of Homeland Security, citing operation of an unlicensed money transmitting business, began seizing funds from Mt.Gox's US subsidiary: by August it had taken about $2.9 million from the Dwolla account and about $2.1 million from a Wells Fargo account.

In hindsight, this was the first official warning signal the company received that year, nine months before it went down.

The same month, CoinLab sued Mt.Gox for $75 million. The legal claim was that Mt.Gox had breached the 2012 "exclusive US market representative" agreement. It went on to become the longest-running claim in the Mt.Gox bankruptcy.

About 630,000 BTC taken between 2011 and 2013

Wizsec's July 2017 report, "Breaking Open the MtGox Case," lays the line out clearly: the theft began in September 2011 and had largely stopped by mid-2013, totalling about 630,000 BTC. On that account the problem did not appear suddenly in February 2014; coins had been draining out since 2011, for nearly two years.

On where the coins went, the Wizsec report states that some 300,000 BTC passed through Vinnik's wallets and ended up on BTC-e, an exchange that, according to the US Justice Department, was run through the shell company Canton Business Corporation. In July 2017 Alexander Vinnik, charged by the US as one of BTC-e's operators, was arrested in Greece (Chapter VII covers the full investigation).

Chapter IV: Transaction malleability — does Mt.Gox's official explanation hold up?

This chapter deals with the explanation Mt.Gox itself gave. The public statement it put out after suspending withdrawals in February 2014 attributed the problem to a protocol-level Bitcoin bug called transaction malleability.

What transaction malleability actually is

Every bitcoin transaction has a unique identifier (txid) computed as the SHA-256 hash of the serialized transaction data. Two identical transactions should produce two identical txids.

However, between 2010 and 2013, the Bitcoin protocol allowed multiple valid signature encodings for the same transaction — that is, the same effective transaction could be represented with several different but equally valid DER-encoded signature byte sequences. Each different encoding produced a different byte sequence overall, and therefore a different SHA-256 hash, and therefore a different txid for what was logically the same transaction.

The implications of this were:

• User A sends 1 BTC to exchange B. The original txid is, say, abc123.
• A third party watching the mempool re-broadcasts the same transaction using a different but valid DER encoding. The new txid is, say, def456.
• Sometimes a miner picks up def456 first and includes it in a block. abc123 is now invalid because its inputs are spent.
• User A queries the exchange asking "did my transaction land?" The exchange's code is naively looking for abc123 on the chain. It does not see abc123. It treats the deposit as failed.

This bug was real. The Bitcoin community had been discussing it since 2011.

What Mt.Gox said, and what that cannot explain

Mt.Gox's February 10, 2014 press release pinned the problem on malleability: in substance, that attackers altered the txids of outgoing withdrawals so that the system believed the withdrawals had failed and sent them again.

The explanation is technically possible. But as a full account of the missing coins it fails to fit in three places:

1. A sound withdrawal system checks whether UTXOs have been spent, not just whether a specific txid landed. UTXOs cannot be malleated — once spent, they are spent.
2. The window for a successful malleability attack is short. From transaction broadcast to confirmation is typically about ten minutes. To extract hundreds of thousands of BTC via malleability alone, the attacker would need to execute the attack tens of thousands of times, which would leave a large trail on chain.
3. The timing does not fit. Wizsec's 2017 investigation showed that about 630,000 BTC were drained from the hot wallet between September 2011 and mid-2013, long before the February 2014 malleability episode. That loss cannot be explained by malleability.

So: the malleability bug was real, but it can account for only a small part of the total loss. It cannot explain the larger fact: the accounting books had not corresponded to chain-side reality since 2011.

If you don't write code, the simplest way to put it: malleability was like a keyhole that sits slightly off-angle — a real but small flaw. It does not explain why the whole house was empty.

How the Bitcoin community later fixed it

The Bitcoin community later pursued two fix paths for malleability:

• BIP62 (proposed 2014) — attempted to restrict signature encoding formats to remove malleability. BIP62 was eventually withdrawn because it could only address third-party malleability, not "sender-initiated" malleability.
• SegWit (activated August 2017) — separated signature data from the main transaction body ("segregated witness"), removing the signature hash from txid calculation entirely. This is the complete fix.

Chapter V: The week the ship went down

On February 7, 2014, Mt.Gox halted all BTC withdrawals. On February 10, the official explanation cited the transaction malleability bug covered in the previous chapter.

The bug was real, and other exchanges faced it in the same window: Bitstamp paused for several days, deployed a patch, and resumed. Mt.Gox paused, and never resumed.

That week an English programmer named Kolin Burges flew from London to Tokyo and stood outside the Shibuya building that housed Mt.Gox with a sign reading "Mt.Gox: where is our money."

March 20, 2014: about 200,000 BTC found in an old wallet

On March 20, 2014, Mt.Gox announced on its website that it had found 199,999.99 BTC in an old-format wallet used before June 2011. Those roughly 200,000 BTC became the funding pool for the later creditor distributions.

Chapter VI: A comparison — BitGrail in 2018

Chapter VII: On-chain forensics — Wizsec, Kim Nilsson, and the BTC-e thread

There is one more thread in the Mt.Gox story: independent investigators, using only public chain data, reconstructed where the stolen BTC went. That thread led to BTC-e; Alexander Vinnik, charged by the US as one of its operators, was arrested in Greece in 2017 and pleaded guilty in the US in 2024 (his French conviction concerned laundering linked to the Locky ransomware, not Mt.Gox).

Who Wizsec are

WizSec is a Tokyo security company. The main investigator on the Mt.Gox coins was the Swedish engineer Kim Nilsson, who wrote WizSec's 2017 write-up. He did not go down the litigation route; he followed the money through public chain data, and the trail ended at BTC-e.

The basic method for this kind of on-chain tracing:

1. Seed set: take the known Mt.Gox wallet addresses as the starting set.
2. Outflow tracing: follow all BTC leaving these addresses to their next-hop destinations.
3. Cluster identification: apply usage-pattern heuristics to the next-hop addresses — for example, the "common-input-ownership" heuristic (when a transaction has multiple inputs they are usually controlled by the same private key); same-timing-window common-script-template outflows (same wallet software, same timing pattern).
4. Exchange-deposit pattern matching: use the resulting clusters to identify "intermediate laundering clusters," then trace those clusters into known exchange deposit addresses.

The July 2017 report

Wizsec's July 2017 report, "Breaking Open the MtGox Case," is the key on-chain forensic document for this event. Its core findings:

• In September 2011 the Mt.Gox hot-wallet private keys were stolen via a copied wallet.dat file; by mid-2013 the thief had taken out about 630,000 BTC.
• Of that, some 300,000 BTC passed through Vinnik's wallets and ended up on BTC-e.

The case against Alexander Vinnik

Alexander Vinnik is a Russian citizen. The 2017 US indictment charged him as one of the operators of BTC-e.

2017.07 — Vinnik is arrested in Greece. The US charges include money laundering and operating an unlicensed money services business, 21 counts in all.

2017-2020 — the US, France, and Russia all seek his extradition from Greece.

2020.01.23 — Vinnik is extradited from Greece to France. In December 2020 a Paris court convicts him of money laundering and sentences him to five years and a €100,000 fine; the extortion charges tied to the Locky ransomware do not stand. The French case concerned laundering linked to Locky, not Mt.Gox.

2022.08 — Vinnik is extradited from Greece to the United States and makes his first appearance in federal court on the 2017 indictment.

2024.05 — Vinnik pleads guilty to conspiracy to commit money laundering in connection with operating BTC-e. The Justice Department holds him responsible for losses of at least $121 million.

2025.02.12 — the US government announces Vinnik's release in exchange for Marc Fogel, an American teacher imprisoned in Russia; Vinnik lands in Moscow the next day.

Chapter VIII: Four viewpoints — user, Karpelès, court, investigator

The previous chapters followed the timeline. The same events look different depending on where you stand, so this chapter goes through them from four positions.

Viewpoint one: the user

Suppose you were a Mt.Gox user. This is how it unfolded:

• February 7 — BTC withdrawals are paused.
• February 10 — Mt.Gox blames the malleability bug, which sounds like a technical issue that will be fixed in a few days.
• February 24 — Mt.Gox suspends all trading and, hours later, the site goes offline. The same day, a "Crisis Strategy Draft" leaks.
• February 28 — the company says about 850,000 BTC are missing.
• Over the following decade — the bankruptcy and rehabilitation proceedings run on.
• July 2024 — base distributions begin. The BTC returned by quantity is worth, at 2024 prices, far more than what was deposited in 2014.

Viewpoint two: Karpelès

Karpelès's account is different. He has never admitted to actively stealing BTC, and he has maintained that most of the loss happened without his knowledge. The Japanese court acquitted him of professional embezzlement (gyōmujō ōryōzai) and convicted him of unauthorized creation of electromagnetic records, that is, falsifying system data.

Viewpoint three: the court and the trustee

The court-side viewpoint runs mostly through Nobuaki Kobayashi, the lawyer who has served as Mt.Gox's trustee since 2014. The case has run since 2014 and is still open.

The main milestones:

• 2014-2015 — the trustee takes over Mt.Gox's assets, including the roughly 200,000 BTC found in the old wallet.
• 2017-2018 — the trustee sells part of the BTC and BCH for yen.
• 2018 — the procedure is reclassified from straight bankruptcy back to civil rehabilitation (民事更生). Under bankruptcy, creditors would have been paid in yen at the 2014 valuation; civil rehabilitation allowed them to be paid in BTC quantity terms.
• 2021 — final rehabilitation plan approved by creditor vote. Two options offered: "early lump-sum" (lower amount, faster) versus "intermediate-stage installments" (higher amount, slower).
• 2024-2026 — execution of Base Repayment.

That creditors got BTC back by quantity at all runs through this legal procedure.

Viewpoint four: the investigator

On the investigator side, Chapter VII has already covered Wizsec and Kim Nilsson. One more person connected to these events is Kraken founder Jesse Powell: after Mt.Gox was hacked for the first time in June 2011, Powell went to Tokyo to help; when Mt.Gox distributions began in 2024, Kraken was one of the exchanges creditors could designate to receive them.

Chapter IX: Twelve years and still not finished

The Mt.Gox bankruptcy proceeding, counting from the April 24, 2014 Tokyo District Court ruling that converted the initial civil rehabilitation filing into formal bankruptcy, has now run for twelve years and is not yet closed. The key milestones:

The 2018 reclassification — the biggest turning point in the procedure

Under straight bankruptcy, creditors would have received yen-denominated payouts calculated against the BTC price as of 2014. Civil rehabilitation, in contrast, allowed the procedure to distribute on BTC quantity terms. The BTC that creditors got back by quantity was, by the time of the 2024 distributions, worth far more in dollar terms than what they had deposited.

Karpelès's Japanese criminal trial (2015-2019)

On March 15, 2019 (Tokyo time) the Tokyo District Court handed down its first-instance ruling (AFP report of that day, carried by phys.org; for the inflated-holdings figure see Wikipedia · Mark Karpelès). The charges included:

• Professional embezzlement (業務上横領罪) — prosecution alleged Karpelès had moved approximately 341 million yen of customer deposits to his personal account.
• Unauthorized creation of electromagnetic records (私電磁的記録不正作出罪) — prosecution alleged Karpelès had falsified data in the Mt.Gox system, inflating its holdings by about $33.5 million.

Verdict:

• Professional embezzlement and the other charges — not guilty.
• Unauthorized creation of electromagnetic records — guilty. Sentence: 2.5 years' imprisonment, suspended for 4 years.

Chapter X: Where the distributions stand (checked October 2026)

The publicly verifiable status is as follows:

• Before base distribution began, trustee Nobuaki Kobayashi held approximately 142,000 BTC; distributions to creditors have been going out since July 2024
• On October 27, 2025, the Rehabilitation Trustee announced that the deadline for base repayment, early lump-sum repayment and intermediate repayment had been moved from October 31, 2025 to October 31, 2026 (Japan time) (trustee's notice of October 27, 2025). For any further change, go by the notices on www.mtgox.com

Before and after every distribution wave, the market routinely worries that the trustee will sell BTC into the market and trigger a sell-off. In practice, the trustee moved the distributed BTC to creditor-designated exchange addresses, and the actual sell decision was left to each individual creditor.

The CoinLab dispute

The longest-running claim against the Mt.Gox estate came from CoinLab — a US company that signed a 2012 "exclusive Mt.Gox US-market agent" contract. In May 2013 CoinLab sued Mt.Gox in the US District Court for the Western District of Washington for $75 million.

On January 15, 2021 Bloomberg reported that CoinLab had reached an agreement with trustee Nobuaki Kobayashi and Fortress's Mt. Gox Investment Fund, under which as much as 90% of the bitcoin remaining in the proceedings would be offered to creditors (Wikipedia · Mt. Gox). The rehabilitation plan was approved by creditors that October.

Chapter XI: What to take away from Mt.Gox

The usual takeaways from Mt.Gox are two: don't leave your keys on an exchange, and CEXes are untrustworthy. The points below are more specific:

1. An exchange with no independent reconciliation may not know its own books are off. About 630,000 BTC were drained from Mt.Gox's hot wallet from September 2011 onward; when the company became aware of it has never been established. When judging a venue, look first for a report issued by an independent third party and for reserves you can cross-check on chain. Lesson Two turns this kind of judgement into five criteria you can check one by one.
2. A regulator's first move is worth noting on its own. US Homeland Security seized Mt.Gox's accounts in May 2013, nine months before it went down. Any exchange you use that receives a subpoena, a fine or a warning from a regulator should be marked specifically.
3. Treat a "withdrawals paused" notice as a sign of serious trouble first. Bitstamp resumed a few days after its 2014 pause; FTX, Celsius, Voyager, BitGrail and QuadrigaCX never resumed after theirs. If you see a withdrawal-pause announcement on an exchange you use, move out whatever you still can as quickly as possible, and be prepared for the rest to be hard to recover.
4. When an exchange's head talks about risk, listen for mechanisms rather than adjectives. Reconciliation frequency, who runs the audit, how often reserves are attested — those are mechanisms; "we place the highest importance on security" is an adjective. Listening is not enough on its own; you also want to see what he can actually produce for verification.
5. Verbal assurances do not count; verifiable evidence does. SBF tweeted on November 7, 2022, "FTX is fine. Assets are fine"; four days later FTX filed for bankruptcy. The way to judge whether an exchange will actually fix a problem is whether it produces evidence you can verify on chain.
6. Keep any single exchange position to an amount you could lose entirely. When small signals such as delayed withdrawals appear, that is the time to start reducing — do not wait for an announcement.