Web3 security is not a subscription, a browser extension, or a checklist you finish. It is a small set of assumptions about how the system works, and the losses come from believing a different set.

Here are the five that matter. If you internalise only these, most of what we document in the incident files becomes predictable: you can see the trap before the page finishes loading.

1. A wallet is a key, not an account

A bank account is a relationship with an institution that can freeze, refund or investigate. A wallet is a number that signs things. There is no institution, no fraud desk, and no identity binding.

That means:

  • Whoever holds the private key or seed phrase is you, as far as the chain is concerned.
  • There is no “recovery” that does not involve someone else holding a copy in advance — a multisig, a social-recovery module, or a custodian you are choosing to trust.
  • The address is not a vault; it is a name that keys can sign for. Moving funds means creating a transaction, not “transferring ownership” in any legal sense.

The practical consequence: your security budget belongs on key custody first, everything else second.

2. Signing is not logging in

The most expensive misunderstanding in crypto. On a normal website, “sign in with your wallet” is meant to prove you control an address. But the same popup can authorise a transfer, approve a contract, or grant a delegated spender.

Wallets try to distinguish these: readable messages, typed data, transaction simulation. Attackers try to blur them. A signature replay attack does not break your key; it reuses a signature you gave once for a second purpose. A permit signature can grant spending rights without a transaction at all.

The habit worth building: before you sign, ask what the signature would let the other side do if they were hostile. If the answer is “move tokens”, the site is asking for custody, not identity.

3. Approvals do not expire

An approval is a standing permission recorded on-chain: this contract may move up to X of my token, until I revoke it. Nothing in the protocol times it out, and nothing notifies you when it is used.

That turns every interaction into a small, permanent delegation. Approvals granted to a protocol that later gets exploited become the attacker’s inventory — the Ankr bridge incident and countless drainer waves both ended by draining tokens that were still approved.

Two habits fix most of this: approve a specific amount rather than “unlimited”, and put a recurring date in your calendar to review the approval dashboard. Our token approval safety guide covers what each permission means in practice.

4. A contract is not its website

The website is marketing; the contract is the code. They can disagree, deliberately.

A smart contract can hold funds, call other contracts, and change behaviour based on inputs you cannot see. It can also be upgradeable, meaning the behaviour you audited last month is not the behaviour running today. Contracts that look like tokens can be honeypots; contracts that look like bridges can accept forged proofs.

You do not need to read Solidity. You need to stop treating a site’s description as evidence about the code, and to notice when a transaction asks for something the site never mentioned. That gap is where most incidents live.

5. There is no undo

Every other model collapses into this one. A signed transaction is final. A leaked key ends the discussion about who owns the balance. A bridge that releases funds to the wrong party will not be talked out of it.

Because there is no undo, the only workable strategy is to reduce the amount that one mistake can move:

  • Separate wallets by purpose, so a mistake hits pocket money.
  • Test transfers before large ones.
  • Prefer protocols with delay windows and pause functions when the amount is serious.
  • Keep the majority of assets behind something that requires two independent approvals — a multisig or a hardware wallet plus a separate signing device.

Where web3 security incidents actually happen

If you map real incidents onto these five models, the distribution is lopsided. Our own archive and the public post-mortems agree: most losses come from signatures and approvals, not from broken cryptography.

  • A drainer convinces you to sign a transfer or a permit. Model 2.
  • An approval left open gets swept when a protocol is exploited. Model 3.
  • A website shows one thing, the contract does another. Model 4.
  • A seed phrase is exposed through a phishing page or a bad backup. Model 1.
  • A bridge accepts a forged receipt and pays out. Model 4, with Model 5 doing the damage.

The fastest way to learn the shape is to read a handful of case files end to end rather than a dozen listicles. Start with how to avoid crypto phishing scams, then bridge security explained, then our wallet security checklist to turn the models into settings.

Limitations

These models explain the losses, not every risk. They do not cover protocol-level economic attacks (oracle manipulation, governance capture), they do not make a bad counterparty safe, and they say nothing about the legal exposure of holding assets in a jurisdiction that decides to freeze them.

They also assume you are the only person with access. If a platform holds your keys, your security model is that platform’s security model plus its incentives — a different conversation, covered in the exchange hacks we have written about.

For everything you actually control, the five models are enough. Almost nobody loses money because the curve was broken.

Frequently Asked Questions

What does web3 security actually mean?

It means assuming the opposite of a bank. You are the custodian, signatures are irreversible contracts, and a website is not the protocol. Web3 security is the practice of reading what you are about to authorise at three layers — key, signature, contract — before the transaction leaves your wallet.

Is web3 less safe than normal banking?

It is differently unsafe. Banks absorb fraud and reverse mistakes; that convenience is priced in and hides how much trust you are placing in them. Web3 removes the intermediary, which removes the undo button. The raw risk is comparable for careful users and much higher for careless ones, because there is no backstop.

What is the most common way people lose funds in web3?

Signatures. Not broken cryptography: a user authorises something they do not understand, on a site they do not control the code for. Phishing pages, malicious token contracts, permit signatures and blind signing all exploit the same gap between what the screen says and what the transaction does.

Do I need to understand smart contracts to be safe?

Not the code, but the behaviour. A contract is a program that can hold your assets and call other programs; it can be honest, buggy, or hostile, and its website tells you nothing about which. Knowing that much is enough to justify the habits that matter: limited approvals, burner wallets, and refusing to interact with tokens you did not choose.

How do I start learning web3 security without a technical background?

In this order: how keys and seed phrases work, what a signature authorises, what an approval is, then how incident cases actually unfold. Reading five real post-mortems teaches more than any theory, because the pattern repeats: one signature, one approval, one compromised key, then irreversible settlement.