$1,000,000 in security audit grants are live now, Apply here →

← Resource Hub

Cross-Chain Bridge Security: How Bridges Get Hacked

By the Guardian team Updated 6 min read

Cross-Chain Bridge Security: How Bridges Get Hacked

Quick answer. Bridges hold pooled funds and trust a few verifiers to report other chains, so one stolen key, faulty message check or weak setting can release them. Independent verifiers, rate limits and guarded upgrade keys limit the damage.

Watch: Cross-Chain Bridge Security: How Bridges Get Hacked 1:41
Video transcript

Bridges hold pooled funds and trust a few verifiers. One stolen key, faulty message check or weak setting can release them.

A bridge acts on a message from another chain. Whoever can forge or sign that message can take funds nothing backs.

Five bridge incidents were valued at more than 1.5 billion dollars when they happened. Each broke one of four links.

Ronin and Multichain centered on keys. Wormhole's signature check was bypassed, and a Nomad upgrade made every message valid. Kelp DAO relied on a single verifier.

Ronin needed five of nine validator signatures. Sky Mavis ran four, and a forgotten allowlist supplied the fifth. Keys are only as independent as the people behind them.

At Kelp DAO, poisoned RPC nodes led its only verifier to confirm transactions that never took place. LayerZero, a Guardian client whose code we audit, says Kelp chose that setup. Kelp says it followed LayerZero's documented defaults.

Require two or more independent verifiers on every route, and never rely on defaults. Rate limits cap a forged release. Guard upgrade keys with a multisig and a timelock.

Guardian audits and pentests bridge contracts, keys and servers. Since February 2025, we have verified every step of USDT0's omnichain rollout. Managed Security then reviews every multisig transaction before signing.

Read the full article for every incident and the full checklist, or talk to Guardian.

A bridge is a vault with a messenger. It releases or mints tokens on one chain when told that something happened on another.

That message is the target. Whoever can forge it, or sign it, can take funds that nothing backs.

The five incidents below were valued at more than $1.5 billion when they happened. Each broke one of four links: keys, message checks, configuration or an upgrade.

Disclosure: LayerZero is a Guardian client, and Guardian audits its code. On Kelp DAO, we give both sides' accounts.

How Bridges Move Value

Two token designs are common.

  • Lock and mint: lock tokens on the source chain, mint wrapped copies on the destination. The lockbox backs every copy.
  • Burn and mint: burn on the source, mint on the destination. No lockbox, so the mint path is the prize.

LayerZero's OFT standard offers both: an OFT burns and mints, while an OFT Adapter locks and unlocks.

Either way, the destination contract acts on a message it cannot check itself. A messaging layer carries it, and verifiers vouch for it.

Who verifies the message

  • Validator sets: a fixed group of keys signs releases. Ronin needed 5 of 9.
  • Signer networks: Wormhole runs 19 nodes, called Guardians, and needs 13 signatures for a valid message.
  • MPC custody: a key shared across nodes controls the bridge addresses, as at Multichain.
  • App-chosen verifiers: on LayerZero, each app sets required and optional verifiers, called DVNs, per route.

A bridge is only as strong as the smallest group that can approve a release.

For how two messaging layers compare, see LayerZero vs Chainlink CCIP.

Five Bridge Losses at a Glance

IncidentDateReported lossWhat broke
WormholeFebruary 2022120,000 wETH, about $326MSignature check bypassed on Solana
RoninMarch 2022About $624MFive of nine validator keys
NomadAugust 2022About $190MAn upgrade made every message valid
MultichainJuly 2023About $125MUnexplained outflows from MPC addresses
Kelp DAOApril 2026About $290MOne verifier fed forged chain data

Losses as reported at the time, before any recovery.

Key and Signer Compromise

Ronin, March 2022

The attacker took 173,600 ETH and 25.5 million USDC, about $624 million, per Rekt.

Withdrawals needed 5 of 9 validator signatures. Sky Mavis ran four of them.

The fifth came from an old favor. In November 2021, Axie DAO let Sky Mavis sign on its behalf during heavy traffic.

That was meant to end the next month, but the allowlist access was never removed.

The theft was found nearly a week later, when a user could not withdraw 5,000 ETH. Sky Mavis then said "the threshold will be eight out of nine."

Multichain, July 2023

On July 6, 2023, about $125 million moved out of Multichain's bridge wallets, DL News reported. Multichain said assets on its MPC address had moved "to an unknown address abnormally."

On July 14, Multichain said its CEO had been in custody since May 21. Multichain also said its access keys to the infrastructure hosting its MPC node servers had been revoked.

Multichain did not say who moved the funds, DL News noted. The team said it was "forced to cease operations."

The lesson in both: distributed keys are only as independent as the people and accounts behind them.

Message Verification Bugs

Wormhole, February 2022

An attacker minted 120,000 wETH on Solana with no ETH behind it, about $326 million, per Rekt.

The Solana contract did not properly check an account address used in its signature verification. With a fake account, the attacker bypassed the check on Wormhole's guardian signatures, then minted.

Wormhole restored the backing within a day: "ETH will be added over the next hours to ensure wETH is backed 1:1."

Nomad, August 2022

About $190 million was drained in roughly two and a half hours, per Rekt.

A June upgrade had initialized Nomad's Replica contract with 0x00 as a trusted root. As Rekt put it, "all messages were read as valid by default."

Others then copied the first exploit transaction. No key was stolen; the check itself was broken.

Both failures sat in the code that decides whether a message is genuine. Review that code, and every upgrade to it, hardest.

Configuration Failures

Kelp DAO, April 2026

On April 18, 2026, Kelp DAO was "exploited for approximately $290M," LayerZero said.

Its rsETH route "relied on a 1-of-1 DVN setup, with LayerZero Labs as the sole verifier."

The attacker compromised two RPC nodes, then hit the healthy ones with a denial of service. That forced failover to the poisoned nodes.

The DVN then "confirmed transactions that never in fact took place." LayerZero says the attacker could not compromise the DVN instances themselves.

The two sides describe fault differently:

  • LayerZero: its protocol "functioned exactly as intended," and Kelp chose 1-of-1 despite recommendations to diversify.
  • Kelp: the setup "followed LayerZero's documented defaults," and the compromised stack was LayerZero's own, per crypto.news.

LayerZero's docs warn that defaults "may include only a single DVN, and may change without notice." Our Kelp DAO breakdown covers the rest.

Settings that act like keys

Verifiers are one setting among several. Peers, confirmations, libraries and executors also decide which messages a route accepts.

On LayerZero, the delegate "independently controls all Endpoint and security configuration, with no signature required from the Owner." Guard it like an upgrade key.

Bridge Security Checklist

Verifiers and keys

  • Require two or more independent verifiers on every route.
  • Count operators, not keys: four of Ronin's nine keys sat with one company.
  • Revoke temporary signing rights the day the need ends.
  • Make sure losing any one person cannot cut you off from signer infrastructure.
  • Give each verifier independent RPC sources, and alert when they disagree.

Configuration

  • Set every route's verifiers, confirmations and libraries explicitly. Never rely on defaults.
  • Check both directions of every route, after every change.
  • Put owner and delegate roles on multisigs, never a single key.
  • Alert on every configuration, peer and delegate change.

Rate limits and pauses

  • Set inbound and outbound limits per destination, sized to normal flow. Inbound limits cap a forged release.
  • Keep a pause per route and a global pause.
  • Split pauser and unpauser, so "a compromised pauser key cannot also undo a legitimate security pause."
  • Reconcile the lockbox against wrapped supply on every chain, and alert on any gap.

Upgrade keys

  • Put upgrades behind a multisig and a timelock.
  • Re-review every upgrade, including its initialization values. Nomad's loss followed a June upgrade.
  • Match deployed bytecode and settings to the reviewed code before a route carries value.

Our LayerZero OFT checklist covers token contracts. Our deployment verification checklist covers each chain's rollout.

How Guardian Secures Bridges

Guardian's audits and pentests cover the contracts and the systems around them. OpSec audits test keys and signers; infrastructure pentests test servers and cloud accounts.

Since February 2025, Guardian has verified every step of USDT0's omnichain rollout, as our case study describes.

That means each chain's peers, DVN sets, confirmation counts and enforced gas, for every pathway. The work spans more than 45 engagements across more than 25 chains.

After launch, Managed Security reviews every multisig transaction before it is signed, and every configuration change as it lands.

Frequently Asked Questions

Why do cross-chain bridges get hacked so often?

They hold pooled tokens, and a small set of keys or verifiers decides when to release them. Steal enough keys, fool the verifiers or break the code that checks messages, and the bridge pays out.

What is the difference between lock-and-mint and burn-and-mint bridges?

A lock-and-mint bridge locks tokens on the source chain and mints wrapped copies, so the lockbox is the target. A burn-and-mint bridge burns and re-mints, so the mint path is the target.

How many verifiers should a bridge route have?

At least two independent operators, each required to approve. LayerZero's docs advise production routes to include at least one required DVN not run by LayerZero Labs. Kelp DAO's exploited route had one verifier.

Do rate limits stop bridge hacks?

No, but they cap the loss. A limit per route and time window bounds what one forged message or stolen key can release, and buys time to pause. Set them on both sides of every route.

Sources

Secure your bridge before it carries value

Guardian reviews your bridge contracts, verifiers, rate limits and upgrade keys, then every change after launch.