Blockchains cannot read each other. Moving an asset from one to another therefore means locking it on the first chain and issuing a representation on the second, with some mechanism deciding when the lock is genuine. That mechanism is the bridge, and it is where the value accumulates: the lock contract holds real assets on behalf of everyone who has ever crossed.
The critical property of the result is often missed. A bridged token is not the original asset. It is a liability of the bridge, denominated in the original asset, and its value depends entirely on the bridge remaining solvent and honest. Two bridged versions of the same coin on the same chain are different assets with different risk, which is why they can and do trade at different prices.
The dominant failure has been the validation layer. Most bridges are secured by a set of external signers who attest that a deposit occurred on the source chain. Compromise enough of those keys and you can mint representations backed by nothing. The Ronin bridge lost the equivalent of hundreds of millions in March 2022 after an attacker obtained a majority of a nine-key validator set, and the theft was not noticed for days. Harmony's Horizon bridge failed the same year through a similar concentration. In both cases the money was protected by a small multisignature threshold while the value behind it had grown enormously.
The second failure is verification logic. Wormhole lost a comparable sum in February 2022 when an attacker forged a signature verification and minted wrapped ether without depositing anything. Nomad's failure in August 2022 was an initialisation error that made every message appear valid, producing an unusual event in which hundreds of ordinary users copied the exploit transaction and drained the contract themselves. These are bugs in the part of the system whose only job is to decide what is real.
The third is economic rather than technical. A bridge that mints against collateral it does not fully hold, or that relies on a liquidity pool to service withdrawals, can be solvent on paper and unable to honour redemptions in a rush — which produces a discount on the bridged asset, which propagates into every lending market that accepted it as collateral.
What distinguishes the more defensible designs is where trust sits. A light-client bridge verifies the source chain's consensus directly in a contract on the destination chain, so its security is the source chain's rather than a committee's; it is expensive and hard to build, which is why it is rare. Optimistic bridges add a challenge window and an honest-watcher assumption. Liquidity networks avoid minting altogether by paying out from pre-funded pools on the destination side, converting custody risk into inventory risk. None removes risk; they relocate it to somewhere the user can inspect. The practical questions before using one are how many keys secure it, who holds them, whether the code has been audited since its last upgrade, and whether the value it holds is proportionate to any of those answers.