Bridges

In brief

Blockchains do not natively share state. Bridges copy meaning across that gap: lock-and-mint, burn-and-mint, liquidity networks, or verified messages. A wrapped asset is a claim on whatever the bridge actually holds and whoever can authorize minting. Canonical rollup bridges inherit more of the rollup’s rules than a third-party lockbox. A multisig or small validator set can amount to effective custody even when a contract holds the assets. Light-client and IBC-style designs try to verify the other chain instead of trusting a separate committee. Assets that show the same ticker can have different risk. Major incidents (Wormhole, 2 February 2022; Ronin, 23 March 2022) show unbacked minting and compromised signer control — not “coins vanishing in transit.”

Why bridges exist

ethereum.org: chains are isolated environments with different rules. Tokens do not move freely. Bridges exist so people can use cheaper execution (an L2), another application ecosystem, or a representation of an asset that is native somewhere else.

That is interoperability, not magic. Each side still has its own ledger. A bridge is a protocol for updating both in a coordinated way — plus whatever humans and keys sit in the middle.

Assets do not “move”

Speak carefully.

• Lock-and-mint. Coins are locked in a contract (or with a custodian) on chain A. A wrapped token is minted on chain B. Unlocking on A requires burning or locking the wrap on B. Chainalysis’s Wormhole write-up describes this collateral pattern for ETH / WeETH.
• Burn-and-mint. Native issuance is destroyed on A and created on B. Circle’s CCTP is a concrete issuer-run example: USDC is burned on the source chain and minted on the destination after Circle’s attestation service signs the message. There is no source-chain lock vault, but the attester and token issuer remain trust dependencies.
• Liquidity / intent networks. A relayer or solver pays you from inventory it already has on B, then settles or rebalances later. Across, for example, has relayers front destination-chain capital and be repaid after optimistic settlement. No new wrap is required for the user’s fast payment; availability, settlement, and relayer inventory still matter.

In every case what you hold on B is a representation or a new payment, not the original UTXO or account balance.

Wrapped and native

Wrapped Bitcoin (WBTC) on Ethereum is an ERC-20. ethereum.org’s bridges page is explicit: it is an Ethereum token, not BTC on Bitcoin. To hold native BTC you have to obtain it on Bitcoin, for example through a bridge or an exchange.

Native USDC on a chain, when Circle actually mints there, is a different claim from bridged USDC locked in a third-party contract. Same ticker, different issuer and failure modes. See Stablecoins in Batch C.

A canonical / official rollup bridge is the pair of contracts the rollup team (or governance) treats as the real lock for that L2. Third-party bridges to the same L2 are extra trust. Batch C’s Layer 2 article makes the same split.

Trusted, federated, verified

ethereum.org’s two-column picture is a start: trusted bridges depend on an operator or committee; trust-minimized ones try to inherit the security of the underlying domains.

In practice there is a spectrum:

• Custodial / exchange. The firm credits an internal balance and pays out on the other chain. You have a counterparty.
• Multisig / federation. N-of-M keys sign deposits and withdrawals. Ronin’s 2022 design, as described by Halborn from Sky Mavis’s then-public account, used nine validators and required five signatures. Four validators were controlled by Sky Mavis; an unrevoked allowlist let compromised Sky Mavis systems obtain the Axie DAO validator’s signature. That is a committee, not Ethereum consensus.
• Light-client / proof-based. A contract on A checks headers or proofs from B (or vice versa). Cosmos IBC is a concrete example: on-chain light clients track counterparty consensus state, while relayers submit packet data and proofs. The relayer transports data; the light client verifies it. This is not “no remaining risk” — client bugs or expiry, bad upgrades, source-chain failure, and relayer liveness still matter.
• Optimistic message passing. A posted claim can be challenged during a window. Security depends on the challenge rules and at least one capable watcher acting in time.
• Rollup canonical bridges. Deposit on L1, mint on L2; withdraw by proving the L2 exit. Optimistic exits wait a challenge period (often about seven days — see Batch C Layer 2).

Messages versus tokens

Token bridging is one application of cross-chain messages: “this lock happened,” “this burn happened,” “call this contract on the other side.” ERC-5164 specifies dispatcher and executor interfaces for cross-EVM execution; it deliberately leaves the transport and its security to the bridge implementation. Its unique message IDs illustrate replay protection. The harder question remains who can establish that the source event is final.

If chain A reorganizes after the bridge already minted on B, the two ledgers disagree. Bridges therefore bake in finality assumptions — wait for Bitcoin work, Ethereum justification/finality, or an L2 challenge window. Waiting less is faster and more dangerous.

Relayers, validators, admins

A relayer carries a proof, signed message, or transaction to the destination. In a light-client system, a permissionless relayer need not be trusted for validity because the destination verifies the proof, although missing relayers can still delay delivery. In other designs the party called a bridge validator may be a key in a separate committee, not a consensus validator of either chain. Upgrade keys, proxy owners, pause roles, and signer-rotation powers can change or stop the bridge. The L2BEAT bridge framework treats validation, liveness, liquidity, permissioned actors, and upgradability as separate questions. Admin risk is not a footnote.

Risks, without a leaderboard

• Smart-contract bugs that mint wraps without a lock (Wormhole, 2 February 2022: about 120,000 wrapped ETH minted on Solana without corresponding collateral, per Chainalysis).
• Key and authorization compromise in a small signer set (Ronin, 23 March 2022: 173,600 ETH and 25.5 million USDC; four Sky Mavis validator keys plus an Axie DAO signature obtained through an unrevoked allowlist; discovered on 29 March).
• Custodial theft or freeze on a trusted bridge.
• Finality / reorg mismatches.
• Replay of an old message if nonces and chain IDs are wrong.
• Issuer risk on the wrapped token (blacklist, pause).
• Liquidity or filler shortages on liquidity / intent routes.

ethereum.org lists contract, operator, censorship, custodial, and user-error risk. L2BEAT’s framework adds a useful discipline: identify the validation mechanism, slow path, available liquidity, permissioned actors, and upgrade path instead of trusting the label “bridge.”

Two incidents, used as lessons

Wormhole (2 February 2022). A Solana-side signature-verification bug let an attacker mint 120,000 Wormhole-wrapped ETH without corresponding ETH deposits. The attacker redeemed 93,750 of it for native ETH on Ethereum; the remainder was swapped on Solana. Jump Crypto supplied 120,000 ETH on 3 February to restore full collateralization (Wormhole’s incident report; Chainalysis). The lesson is mint authority: the wrap is only as good as the check that the lock occurred.

Ronin (23 March 2022; discovered 29 March). Not a clever AMM trick. The attacker compromised four Sky Mavis validator keys and used an unrevoked allowlist to obtain the Axie DAO validator’s signature. Five valid signatures met Ronin’s five-of-nine threshold and authorized fake withdrawals from the lock contract. The lesson is committee concentration, authorization hygiene, and monitoring — the theft sat unnoticed until a user could not withdraw (Halborn; contemporary CNBC reporting).

Loss figures in dollars move with price. Prefer token amounts.

What this article is not saying

Bridges are not all decentralized. Using one is not “moving ETH onto Solana” in the physical sense. A canonical L2 bridge is not the same as a third-party lock. No bridge is recommended.