[Advanced] — how assets move between chains, why bridges hold the worst security record in the field, and how to evaluate one.
The kid version first
Your ETH exists on Ethereum. It cannot leave. There is no mechanism by which an Ethereum coin appears on another chain — the chains don’t know each other exists.
So bridges fake it: lock your ETH on Ethereum, and mint an IOU on the other chain. The IOU is worth something only because someone is holding the real ETH and promises to give it back.
That “someone” is the bridge, and it is holding everyone’s money in one place.
The mechanisms
Lock-and-mint. Deposit on chain A into a bridge contract; a wrapped token is minted on chain B. Burn the wrapper to unlock the original. The bridge contract accumulates every asset ever bridged — one contract, enormous value, permanent target.
Burn-and-mint. For natively multichain tokens, burn on A, mint on B. No honeypot, because there’s no locked pool — but it requires the token issuer to control both sides. Circle’s CCTP for USDC works this way, and it’s structurally safer.
Liquidity networks. Pools on both sides; you sell into A’s pool and buy from B’s. No wrapped assets, no honeypot, and capital efficiency plus slippage as the costs. Across and Hop work like this.
Native / canonical bridges. Rollups have one built in, secured by the L1 itself. Slower (a 7-day window for optimistic rollups) and strictly the most secure option → rollups.
The hard problem
A blockchain cannot observe another blockchain. It has no network access and no way to verify a foreign chain’s state. So every bridge needs someone to tell chain B what happened on chain A — and the security of the bridge is exactly the security of that messenger.
The options, in descending order of trust required:
| Verification | How | Trust |
|---|---|---|
| Externally verified | A multisig or committee attests | Trust the committee. Most bridges. Fastest, cheapest, worst |
| Optimistically verified | Assume valid, allow challenges | Trust ≥1 honest watcher |
| Natively / light-client verified | Chain B verifies chain A’s consensus directly, in a contract | Trust only the chains. Correct, expensive, hard |
| ZK light client | Verify a proof of A’s consensus | The endgame. Cheap verification, mathematically sound → zk |
The overwhelming majority of deployed bridges are externally verified, because it’s the only one that was easy to build. That is the whole story of the losses below.
The losses
Bridges hold the worst record in the field, and it isn’t close:
| Bridge | Date | Loss | Cause |
|---|---|---|---|
| Ronin | Mar 2022 | ~$625M | 5 of 9 validator keys compromised — 4 held by one entity, plus stale delegated access |
| Poly Network | Aug 2021 | ~$611M | A function let the caller change the keeper. Funds were returned |
| Wormhole | Feb 2022 | ~$326M | Forged guardian signature via a verification flaw |
| Nomad | Aug 2022 | ~$190M | Upgrade set a trusted root to zero — every message valid. Crowd-drained |
| Harmony Horizon | Jun 2022 | ~$100M | 2-of-5 multisig, both keys compromised |
Add these up and bridges account for the majority of all value stolen in the field’s history.
Three reasons this is structural rather than bad luck:
- Concentration. All bridged value sits in one contract. Maximum reward for one exploit
- Complexity. Two chains, two codebases, cryptographic message verification, relayers, upgrade paths. Enormous surface
- Weak verification. A small multisig protecting hundreds of millions is an economically irrational amount of value behind a handful of keys → key management
Messaging protocols
The generalisation: pass arbitrary messages, not just assets — so a contract on one chain can call one on another. LayerZero, Wormhole, Axelar, Chainlink CCIP, Hyperlane.
Same trust question, harder to see. “Configurable security” and “decentralised verifier network” typically mean a set of nodes attesting, which is the externally-verified model with better branding. Find out who signs and how many are needed — that’s the security model, regardless of what the documentation calls it.
Evaluating a bridge
- Who verifies messages? If the answer is a multisig or committee, that’s your trust assumption. Everything else is secondary
- How many signers, held by how many independent parties? Ronin was 5-of-9 on paper and effectively 1-of-1
- Who can upgrade it, and is there a timelock?
- How much value is locked? That’s the bounty
- Is there a canonical alternative? For rollups, use the native bridge unless the 7-day wait is genuinely unacceptable
- Is the wrapped asset the canonical one? Multiple incompatible wrapped versions of the same token on one chain is common, and holding the wrong one means holding an illiquid IOU
Practical advice
- Use canonical/native bridges where they exist. Slower and much safer
- Don’t leave assets bridged longer than you need. Every day is exposure
- Prefer CCTP for USDC — burn-and-mint, no honeypot
- Bridge small first. Always
- Treat any wrapped asset as a claim on a bridge, not as the underlying asset. If the bridge dies, the wrapper goes to zero while the original is still locked and unreachable
Key insight
A bridge is a claim that something happened on a chain you cannot see, and the security of every bridged asset is the security of whoever makes that claim. Most bridges answer this with a multisig, which is why they have lost more money than every other category combined. The real fix — light clients verifying consensus directly, made affordable by zk proofs — is finally being built, roughly a decade after the problem appeared.
Related
- case studies — Ronin, Wormhole, Nomad in full
- zero-knowledge proofs — the endgame for verification
- rollups — native bridges
- key management
Source: [reference] — Aug 2026.
Software engineering
Infrastructure & security
Computer science
Hardware & engineering