Six Confirmations Is a Price Tag, Not a Receipt: What a 51% Attack Actually Buys

In May 2018 an attacker double-spent roughly $18 million of Bitcoin Gold with rented hashpower. In August 2020 an attacker rewrote about 3,693 blocks of Ethereum Classic for an estimated $5.6 million — spending about $204,000 on the rental. Confirmations were never a guarantee; they were a bill.

Confessions of a Guy Who Got It Wrong

Years ago I sat across from a friend who was about to move a genuinely uncomfortable sum of money into an exchange, and I told him: “Wait for six confirmations. After six, it’s settled.”

I said it with the calm authority of a man who has read a lot of documentation and understood slightly less. What I should have said: after six confirmations, undoing that transaction costs more than most people are willing to pay. Not the same sentence. I have been quietly annoyed with myself ever since, and this piece is the long-form apology.

The mistake is not exotic. It is the model imported from traditional finance, where finality is granted by an institution: the bank says the wire is final, and a regulator stands behind it. There is no such referee here. Nobody in this system can declare a transfer finished; all we have is a cost curve.

WHAT YOU WERE TOLD / WHAT THE NODE ACTUALLY GUARANTEES

You think

Six confirmations means the payment is settled.

The node thinks

Six confirmations means undoing this costs somebody a lot.

Six confirmations is a settling time in the way a hotel minibar is a security system: nothing physically stops you, there is just an invoice waiting. Everything in this piece follows from treating confirmations as a price rather than a permit.

So let me say the one thing worth taking away, early, before we get to the machinery:

Confirmations do not make a transaction irreversible. Confirmations put a price on reversing it.

Everything below is a corollary of that sentence. Notice how many things you believed about “finality” were really statements about price.

What the White Paper Actually Promised

Go back to page 3 of the original Bitcoin paper and read the sentence people quote without reading:

“If a majority of CPU power is controlled by honest nodes, the honest chain will grow the fastest and outpace any competing chains.”

That is the whole security model, and it is more modest than “immutable” suggests. Three things hide in plain sight:

  • It is conditional. If a majority is honest. Not “assuming cryptography holds” — a wager about the economics of whoever holds the hardware this week.
  • It describes a race, not a wall. “Grow the fastest and outpace”: the honest chain wins by out-running a competitor, with a probability that improves the longer you wait. Probabilistic settlement never converges on “impossible,” only on “absurdly expensive.”
  • The resource is economic. CPU power, later GPU power, later ASIC power: something you can buy, build, or rent by the hour. The security assumption is denominated in money, so the attack surface is too.

Nowhere does the paper say “final,” or “irreversible,” or “settled.” I have read it a dozen times and still import those words from my banking app. The white paper promised a chain that is expensive to rewrite, not one that cannot be rewritten.

Your Chain Is a Working Draft, Not a Stone Tablet

First analogy, and carry it out of this article: the blockchain is not a stone tablet. It is a working draft on a shared desk — any day somebody walks in, drops a heavier stack on top, and the room starts reading the new stack.

That is not a metaphor for “an attack.” It is the designed behavior of every full node. One rule: the chain with the most accumulated work is the truth. No age limit, no burial depth, no point at which a block becomes sacred.

if candidate.accumulated_work > current_tip.accumulated_work:
    reorganize(candidate)   # undo every block since the fork point
else:
    ignore(candidate)       # orphan the challenger

Read the first branch again. reorganize means: roll back every block since the fork point, drop their transactions out of the settled state, dump them back into the mempool, and apply the challenger’s blocks. The blocks you watched get six confirmations are gone — not “contested,” gone, replaced by a history where those transactions simply do not appear.

This is why “blocks buried in concrete” is actively harmful. There is no concrete — a desk, a stack of drafts, and a rule saying the heaviest stack wins. For how that stack is produced, How Bitcoin Mining Works is the longer walk-through; here the point is only that work is measurable, comparable, and purchasable.

Most exposed are light clients. A wallet that only checks Merkle proofs against block headers trusts the header chain — exactly what gets swapped in a reorganization. It cannot see the competing stack arrive. Your own full node can, and will, and will say nothing.

Confirmations Are a Price Tag, Not a Lock

Second analogy, the one that does the real work: confirmations are not a lock, they are a price tag. They work like the minibar in a hotel room.

Nobody stops you eating the chocolate. No lock on the fridge; just a card saying $12. The minibar is not a security system, it is a pricing system: a prohibition converted into an invoice, high enough that most guests decline.

Confirmations are the same object. To rewrite N blocks, an attacker must re-mine them privately and keep out-pacing the honest chain. Every extra confirmation is another hour of global hashing they must overpower. So a confirmation count is not a permission check; it is the number of hours of honest mining someone must buy to un-write your transaction. Six confirmations does not mean “locked.” It means “the invoice is roughly six hours of world hashrate.”

Now the part that turns trivia into risk: the price tag and the prize are set by two different markets. The price comes from the hashrate rental market; the prize from whatever the transaction is worth. When the prize dwarfs the price, the chocolate gets eaten.

A double spend against an exchange is the clean version: deposit, wait for credit, withdraw something liquid, then erase the deposit by publishing a heavier stack in which it never happened. The attacker is not breaking anything. They are paying an invoice — and did the arithmetic before you did.

Six Confirmations Is a Price Tag, Not a Receipt: What a 51% Attack Actually Buys
Foto3821, CC0, via Wikimedia Commons

What a Majority Attacker Can Actually Do

Let me be concrete, because headlines use “51% attack” to mean “blockchain broken,” which is useless. An attacker holding most of the hashing power for a while can:

  • Double-spend their own coins. Their own — coins they already controlled and already sent. This is the money-making move, and the one actually run in the wild.
  • Censor or omit transactions. If you are writing the winning version of history, you can decline to include one you dislike. Not for a moment — for as long as you hold the majority.
  • Orphan other miners’ blocks. Competing blocks get discarded, honest miners lose their rewards, and the attacker collects a larger share of the block subsidy while the majority lasts.

That is essentially the menu. It is short, and note how much of it concerns the attacker’s own activity, not yours.

What They Cannot Do, and Why Everyone Gets This Wrong

This is the section I care about, because the popular imagination of a 51% attack is a hacker vaulting the wall and walking off with your wallet. It is not. A majority attacker cannot:

  • Spend coins whose private keys they do not hold. Not at 51%, not at 90%, not at 99%. Signature verification is not a vote. If it were, the whole system would be worthless.
  • Mint coins beyond the block reward schedule. Every node enforces the issuance curve independently. A heavier stack that pays itself a million extra coins is not a longer chain, it is an invalid one — discarded without its work being counted at all.
  • Change the consensus rules. Not the 21 million cap, not the block size. Those live in validation code on thousands of machines you do not own.
  • Roll back somebody else’s transactions while preserving the rest. Yes, they can rewind history — but everything after the fork point, including their own. They cannot surgically delete your rent payment.

A miner with more than half the hashing power cannot change the system’s rules and cannot take other people’s funds. What they can do is rewrite the existing chain in limited ways: they cannot conjure transactions that never happened, but they can omit any transaction. — Emin Gün Sirer, Cornell

So, plainly: a 51% attack is a rewrite, not a robbery. It does not let an attacker take your coins; it lets a payer un-pay themselves. Your coins under your keys are not the target. The target is whoever already shipped something real against a payment about to be edited out of the record — which is also why the UTXO versus account model debate matters: a reorg rewrites a claim on state, and your model decides how legible that claim is.

One honest caveat, because I dislike articles that hide their own edges: I am skipping selfish mining and block withholding here. They matter in an uncomfortable direction — they let an attacker get leverage with meaningfully less than 51%. They change the cost curve, not the argument: rewriting gets cheaper, the prize does not.

Six Confirmations Is a Price Tag, Not a Receipt: What a 51% Attack Actually Buys
DemieK07, CC BY 4.0, via Wikimedia Commons

Three Times the Bill Actually Arrived

Three real cases, all the same recipe:

ChainWhenWhat was rewrittenDouble-spentAftermath
Bitcoin Gold16–19 May 2018attacker rented cloud hashratea wallet credited with 388,000+ BTG (~$17.5–18M at the time)Bittrex, Binance, Bithumb, Bitfinex, Bitinka hit; Bittrex’s 12,372 BTG compensation demand refused, BTG delisted
Ethereum Classic5 Jan 201915 reorganizations, 12 of them with double spends (counted by 7 Jan)219,500 ETC (~$1.1M)Coinbase paused ETC; deposits and withdrawals resumed only on 11 Mar 2019, requiring a large number of confirmations
Ethereum ClassicAug 2020three deep reorgs in two weeks: ~3,693 blocks (1 Aug, ~12 hours of history), ~4,000 (6 Aug), ~7,000 (29 Aug, deepest)807,260 ETC (~$5.6M) on 1 Aug aloneexchanges pushed ETC confirmation requirements above 10,000

Bitcoin Gold is the template: rent hashrate by the hour, hit deposits, withdraw, un-write them. The 2019 ETC case convinced me the industry took this too lightly — the deep reorg was caught by an exchange’s own monitoring, not the protocol, and the response was to pause a market for two months.

August 2020 is the one I return to, because of one number. Per Bitquery, as reported by CoinDesk, the attacker spent roughly $204,000 renting hashrate and took about 807,260 ETC in the first reorg alone — around 27x. That is not a hack. It is a business with positive expected value, run by someone who read the price tag and found it cheap.

Ten thousand confirmations is not a fix, by the way. It is a refusal to transact, dressed up as a policy.

MESS, and the Day the Hashrate Showed Up

Ethereum Classic did try to fix this at the protocol layer. ECBP-1100, known as MESS, landed in October 2020 and made heavy reorganisations costly on purpose: a competing branch has to show exponentially more verified work the deeper it reaches, so a one-block race is untouched while a 3,000-block rewrite becomes irrational. No checkpoints, no committee — just a fork-choice rule that charges by depth. Clever, and I have a soft spot for it.

Then September 2022 happened. Ethereum moved to proof of stake, its miners had nowhere to go, and much of that hardware landed on ETC — same algorithm. ETC’s hashrate went from roughly 25 TH/s to over 200 TH/s. MESS was subsequently abandoned. It was no longer needed.

Sit with that — it is the most uncomfortable fact here: what actually secured ETC was not a cleverer fork-choice rule. It was an accident — somebody else’s upgrade dumping hashrate onto a chain that happened to speak the same algorithm. Nobody designed that defense. It arrived.

This is also why I refuse to quote “how much it would cost to attack Bitcoin” figures. I have not verified the hardware, facility, or power breakdowns behind them, and I will not launder someone else’s spreadsheet into a fact. The claim I will stand behind is qualitative: the scale of Bitcoin’s SHA-256 hashrate makes renting enough of it to rewrite deep history impractical, simply because no spot market is deep enough to sell you that much. And ETC being cheap to attack in 2020 has nothing to do with market-cap ranking and everything to do with how much hashing power pointed at it that week.

Tier the wait by the size of the check

Six confirmations is an answer for coffee, not for a house. Pick thresholds by amount: small transfers settle in minutes, seven-figure ones wait hours — one number cannot price both.

Small chains share big chains’ algorithms

Hashrate is rented, not bought. A small chain using a popular algorithm can be attacked by the hour, so its safety depends on the cost of eating it, not its market-cap rank.

Watch hashrate, not the post-mortem

Track hashrate and reorg depth while it happens. By the time a forensic write-up is published, the double spend has already been withdrawn somewhere else and laundered.

What I Actually Do Now

I no longer tell people “six confirmations means it’s settled.” What I tell them, and what I do myself, comes down to this:

  • Price the reversal, then set the wait. Do not pick one confirmation number forever. Ask what it costs to rewrite the last six hours of this chain, then compare that to the size of the transfer. If the transfer is worth more than the rewrite, wait longer. One number for every amount is not a policy, it is a habit.
  • Treat algorithm-sharing small chains as a different asset class. Two chains on the same algorithm means one rental market can attack both, and the smaller one is the cheaper target by definition. This is not about which team shipped better code; it is arithmetic about where rented hardware goes.
  • Watch the hashrate, not the news. The exchanges that got hurt found out because they watched their own chain. I would rather see a hashrate spike in real time than read an excellent post-mortem three days after my coins left. Monitoring is boring, and it is the only item here that fires early.

And the correction I owe my friend, said properly at last: your coins are safe from a 51% attack in a way your payments are not. The attacker cannot open your wallet; they can only make a payment they sent you stop having happened. Which is why the losers are never the holders — they handed over something real for a transaction that was still, at that moment, only a price tag.

(The End)

Blockchain

Aave Wants Native Bitcoin Collateral. The Trust Moved, It Didn't Disappear.

2026-10-16 21:51:42

Blockchain

Multisig Wallets, Read From the Source: How 2-of-3 Fails and What It Still Buys You

2026-10-17 13:25:32

❯
Profile
Cart
Coupons
Search