What is a blockchain, actually?

Most of the money left in about twenty minutes, by one count

October 29, 2026

Starting at 20:05 UTC, by one tracking firm's count, most of the money left in about twenty minutes, across ten blockchains. Another firm describes a few hours across 11. Attackers took more than $6 billion in crypto from the exchange Binance. Source Two days later the FBI said North Korea was responsible and asked companies to block transactions linked to the attackers' wallet addresses. Source North Korea denies it. Source

Your first question is probably the right one. You've heard a blockchain can't be hacked. So what got hacked?

It starts with a fingerprint

A blockchain is built on a hash function. Bitcoin's is called SHA-256. It takes any text, from one letter to a whole book, and gives back a fingerprint of exactly 64 characters. Source

The fingerprint of hello starts 2cf24dba. The fingerprint of hellp starts fdd7585e. They share almost nothing.

You can't work backward from a fingerprint to a long, unguessable text. Finding two texts with the same fingerprint is out of reach today, though "none has been found" is a smaller claim than "none exists." A hash uses no key and no signature, so it isn't encryption, and it doesn't say who wrote anything.

Each block carries the fingerprint of the one before

A block is a batch of records. It also holds the fingerprint of the block before it. That's the whole chain. Source

You can check this on the real thing. Bitcoin's first block, dated January 3, 2009, has a fingerprint starting 000000000019d668. The next block has that exact value written inside it. Source

So change one old record, and that block's fingerprint changes. Now the next block points at a fingerprint that no longer exists. Repair that one, and you break the one after it, all the way to today.

Mine a block, then break it

These are real SHA-256 fingerprints, made on your device. To make the rule harder to fake, a block here only counts if its fingerprint starts with 000. The only way to find one is to keep guessing a number called the nonce. That's mining.

Why it's expensive to rewrite

You needed 3 zeros. Each extra zero multiplies the guessing by 16. A real Bitcoin block from November 3, 2026 starts with 20 zeros. The rule then worked out to between 19 and 20 zeros, or about 6.4 × 10²³ guesses on average. Checking one takes a single hash. Source

Computers on the network follow the chain with the most total work behind it. To change an old block, you must redo it and every block after it, then pass everyone else, who keep going. Source

That's also how it stops someone spending a coin twice. All transactions are announced, and everyone agrees on one history. Say a seller waits for 6 more blocks before trusting a payment. If you held 10% of the network's power and had been racing in secret the whole time, your odds of catching up would be about 0.024%. Source

Not every chain burns electricity

What you just did is proof of work. It spends something outside the system: electricity. Cambridge's best guess for Bitcoin in early November 2026 was about 167 terawatt-hours a year, with a range of 86 to 321, because the estimate depends on assumptions. A terawatt-hour is a billion kilowatt-hours, the unit on a home electric bill, and the figure moves daily. Source

Proof of stake puts something inside the system at risk. On Ethereum since September 15, 2022, a validator (a computer that helps confirm blocks) locks up a deposit of 32 ether, Ethereum's coin, and can lose it for breaking the rules. Source

That switch cut Ethereum's energy use by more than 99.9%. Source It did not make fees cheaper. Source These are two designs with different tradeoffs, and people argue about which is better.

So what has actually gone wrong?

Sort the failures by layer.

• The chain's rules: a 2010 bug briefly created about 184 billion bitcoins, fixed within hours. Source Smaller chains have been rewritten, some by attackers. MIT's tracker saw over 40 rewrites of 6+ blocks in 2019 and 2020. Source
• Code on top: in 2016 an attacker used a flaw to divert about 3.6 million ether from The DAO. Source The US Attorney calls the Terraform case a $40 billion fraud. Source
• Companies and keys: Mt. Gox reported about 650,000 bitcoins lost. Source The US Attorney says FTX's founder stole over $8 billion from customers. Source The FBI blames North Korea for about $1.5 billion taken from Bybit. Source

A private key is the secret that lets you spend. Steal it, and the chain will obey you. At Binance, the company first said it found no sign its keys leaked. Source Its initial report, which outside experts will review, says attackers reached an internal service that signs wallet transactions. Source One tracking firm thinks they made Binance's systems sign instead of stealing keys. Source

What a blockchain can't do

A block saying "Ana owns the moon" passes every check. A chain can show that a record was written and hasn't changed. It can't show the record is true. NIST calls this the Oracle Problem, since a sensor can break and a person can type anything. Source

It also isn't anonymous, exactly. The chain holds no names, but its record is public, and real-world links come from outside it, like an address someone posted. Source It's not fully traceable either: mixers can make money harder to follow. Source

And "nobody can change anything" has limits. After The DAO, a hard fork, a rule change that splits the chain, moved the funds to a recovery address. In the Binance theft, companies including Tether and Circle, which issue stablecoins (tokens meant to hold a steady value), froze billions in them. Source Binance says about $2.85 billion was frozen or intercepted. Source Bitcoin itself can't be frozen. Source

How decentralized is it, and what is it good for?

Over the past year, the top 3 Bitcoin mining pools found 59% of blocks, and the top 6 found 82%. Source A pool often doesn't own the machines in it, and miners can leave one. NIST calls "no one controls a blockchain" a misconception.

It's also slow. Bitcoin confirms about 7 to 9 transactions a second, and Visa averages about 8,200. Source A US government flowchart in the same NIST report asks six questions. Answer no to any and it points elsewhere, usually to a database.

Real uses are hard to measure. The World Food Programme reports using a chain to coordinate aid for 6.3 million households, by its own count. Source The European Central Bank launched a system in September 2026 linking blockchain-style platforms to its payment services. It's too new to show how much it will be used. Source About 80% of stablecoin transactions are by bots, so activity isn't the same as usefulness. Source

We explain how this works. We never say what to buy, and this lesson uses no real money or wallet.

Your toy and the real thing

So you don't walk away with a wrong picture, here's where the version you just mined differs from Bitcoin.

• Your Block 2 updated its copy of Block 1's fingerprint for you. A real chain stores it, so the mismatch stays until someone repairs it.
• Bitcoin hashes twice. Yours hashes once.
• It hashes an 80-byte binary header. Yours hashes text.
• It compares a number to a target. Yours counts zeros.
• A real miner runs out of nonces and has to change something else too. Yours never does.

The idea is the same. The scale isn't: compare your 3 zeros with 20.

Build a blockchain, then attack it

• You will have A working mini blockchain you wrote in Python, plus proof of how it can be fooled.
• Time About 60 minutes.
• You need A computer and JupyterLite (free, no account, runs in your browser). It uses no real money, no wallet and no blockchain network. The "points" are just text.

import hashlib, json
FIELDS = ('index', 'data', 'prev', 'nonce')

def block_hash(b):
    text = json.dumps({k: b[k] for k in FIELDS}, sort_keys=True)
    return hashlib.sha256(text.encode()).hexdigest()

def mine(index, data, prev, difficulty=3):
    b = {'index': index, 'data': data, 'prev': prev, 'nonce': 0}
    while True:
        h = block_hash(b)
        if h.startswith('0' * difficulty):
            b['hash'] = h
            return b
        b['nonce'] += 1

def valid(chain, difficulty=3):
    for i, b in enumerate(chain):
        if block_hash(b) != b['hash'] or not b['hash'].startswith('0' * difficulty):
            return False, i
        if i and b['prev'] != chain[i-1]['hash']:
            return False, i
    return True, None

1. Build. Mine a first block with data 'start' and prev '0', then add two "Ana pays Ben 5 points" blocks, using each block's hash as the next one's prev. valid(chain) should say (True, None).
2. Lie well. Add a block saying "Ana owns the moon." Does it pass?
3. Tamper. Change the data in the block at index 1. It should now say (False, 1).
4. Cheat harder. Re-mine that block and everything after it in your own copy. What does valid say now? What does that tell you about one copy checking itself?
5. Compare with a friend. Build an honest copy and compare the last hashes. That's why many copies matter.
6. Feel the wait. Try difficulty 3, 4 and 5, and time them. Copy your code somewhere before you close the tab.

First, you might like to see Anders Brownworth's demo, where a change turns the later blocks red.

The Binance transfers are still there

They are on the chains. Anyone can look them up, and they should read the same next year.

They don't say who was at the keyboard.

Back to the Lab

Last checked November 4, 2026. Torah Tech researched and wrote this lesson with the help of Claude, an AI made by Anthropic. Any mistakes are ours.