TL;DR
A canonical bridge is the chain's own official bridge, run by the chain itself, moving assets using that chain's native messaging. A third-party bridge is anything else: an independent protocol that moves value between chains it doesn't operate.
Canonical bridges are the most trust-minimized way to move a chain's own canonical asset to and from its settlement layer. The cost: withdrawing from an optimistic rollup back to Ethereum sits through a fault-proof challenge window measured in days.
Third-party bridges split into two families. Lock-and-mint bridges hold your asset and issue a wrapped IOU on the other side, the design behind most large bridge hacks. Intent-based bridges have a relayer front you the real asset in seconds, then verify settlement afterward.
Across is the intent-based example. A relayer fills your order in roughly 2 seconds, then gets reimbursed through background settlement secured by UMA's Optimistic Oracle. It routes across 20+ chains and can move funds L2-to-L2 directly, which canonical bridges generally can't.
For most real transfers, a fast verified third-party route wins. For moving a chain's own canonical asset with no time pressure, the official bridge is the conservative pick.
A canonical bridge is run by the chain it serves. A third-party bridge is run by someone else. Who operates the bridge decides almost everything else about how it behaves.
Deposit ETH into Base or Optimism through its official bridge and you are using a canonical bridge. The chain's own contracts lock your ETH on Ethereum and credit it on the L2 using the rollup's native messaging layer, the same messaging the chain uses to post its state back to Ethereum. No separate company. No pooled liquidity. No wrapped token standing in for the real one. The asset that arrives is the chain's canonical version of ETH, because the chain itself minted it. That is about as trust-minimized as moving an asset to a rollup gets.
The catch shows up on the way back.
A Canonical Bridge Is the Official Route, and It Inherits the Chain's Settlement Timing
Optimism and Base are OP Stack rollups. They use ETH for gas, settle to Ethereum, and secure withdrawals with permissionless fault proofs. Withdraw from the L2 back to Ethereum through the canonical bridge and your funds do not finalize right away. The withdrawal enters a challenge window during which anyone can submit a fault proof against an invalid state. Only after that window closes can you claim on Ethereum.
That window runs days, not minutes. It is not a bug. The delay is the security. It is the time the system reserves for an honest party to prove fraud before money leaves the rollup. A treasury rotating a large position back to mainnet with no deadline can absorb that wait without much cost. Someone trying to move funds and use them the same afternoon hits a wall.
Canonical bridges also move in one shape. They connect a chain to its settlement layer, almost always Ethereum. Going from one L2 to another through canonical infrastructure generally means withdrawing down to Ethereum first, then depositing up into the second chain, inheriting the challenge window on the way down. There is no direct canonical lane from Base to Arbitrum.
Third-Party Bridges Split Into Two Very Different Designs
"Third-party bridge" is a category, not a mechanism, and the two designs inside it carry very different risk.
The older design is lock-and-mint. You send your asset to the bridge, it locks that asset in a contract on the source chain, and it mints a wrapped representation on the destination chain. The wrapped token is an IOU. It holds its value only as long as the locked collateral behind it stays safe. Those locked pools hold large balances and redeem against wrapped tokens, which is exactly what attackers have gone after. The bridge category has lost enormous sums to hacks, and the losses concentrate in this pooled-liquidity design.
The newer design is intent-based. You sign an intent describing the outcome you want: the asset and amount you want on the destination chain. A relayer sees it and fronts you that asset out of its own inventory, usually within seconds. You are done. The relayer gets reimbursed later through a settlement process that verifies the fill actually happened. Your funds never sit in a shared honeypot waiting to be drained, because the relayer is spending its own capital and getting paid back after the fact.
(Native USDC is a third rail worth naming: Circle's CCTP burns USDC on the source chain and mints it on the destination, so you receive canonical USDC rather than a wrapped version. Across routes USDC over CCTP for exactly this reason.)
Across Is the Intent-Based Case, Built for Speed and Any-to-Any Routing
Across has been running intent-based bridging since 2021. A relayer fills your order in about 2 seconds, then gets reimbursed through background settlement you never have to wait on. That settlement is secured by UMA's Optimistic Oracle, which runs on a 1-of-N honest actor assumption: one honest participant is enough to dispute a bad settlement proposal and keep the system correct. Across is built by Risk Labs, the foundation behind UMA, and routes value across 20+ chains using three mechanisms under the hood: its own intents, Circle's CCTP for native USDC, and OFT for tokens that use it.
Two practical advantages fall out of this design:
Across can move funds directly between two L2s. A Base-to-Arbitrum transfer is one intent filled in seconds, not a round trip down to Ethereum and back up through canonical contracts.
The user experience does not inherit any rollup's withdrawal window. The relayer carries the timing risk and the capital; you get the destination asset while settlement happens behind you.
For a third-party bridge, the security history is the line that matters, and it is short to state: Across has processed billions in volume since 2021 with no user funds ever lost. That record, more than any single design claim, is what separates a verified intent route from the lock-and-mint pools behind most bridge losses. You can see how the verification model works in the Across docs.
Which One to Use Comes Down to the Job
Match the bridge to the task, not to a general preference.
Use the canonical bridge when you are moving a chain's own canonical asset to or from its settlement layer and you have no time pressure. Depositing ETH onto an L2 for the first time, or rotating a large balance back to Ethereum where a multi-day wait costs you nothing, is what the official bridge is built for. You give up speed and direct L2-to-L2 routing. In exchange you get the chain's own settlement guarantees with no third party in the path.
Use an intent-based third-party bridge for almost everything else. Moving between two L2s, spending funds the same day, routing across a chain pair the canonical bridge doesn't connect directly, not wanting to wait out a challenge window: all of it points to a fast verified route. The real question is never canonical versus third-party in the abstract. It is whether the transfer in front of you needs settlement-layer finality or needs to be done in seconds.
Most transfers need it done in seconds. Official and best are not always the same bridge.