Across: cross-chain bridges rating breakdown
The best of the third-party designs, and for a structural reason worth understanding. There is no validator committee holding your money. A relayer fronts your funds on the destination chain out of their own capital and is repaid later, and the repayment claim is settled optimistically with a bond. So the seven-day settlement risk is real — it is just that the relayer is eating it, not you. That is a genuinely better transfer of risk than handing custody to a committee of strangers. No exploit to date.
The best of the third-party designs, and for a structural reason worth understanding. There is no validator committee holding your money. A relayer fronts your funds on the destination chain out of their own capital and is repaid later, and the repayment claim is settled optimistically with a bond.
How the score is built
Each criterion is scored 0–10 and weighted. The median column is the middle score across every entry in this ranking, so a row reads as a position rather than a number.
| Criterion | What we measured | Weight | Score | Category median | Rank | Weighted gap |
|---|---|---|---|---|---|---|
| Trust assumption | — | 40% | 8 | 4 | 3 of 16 | +1.60 |
| The upgrade key | — | 15% | 5 | 3 | 3 of 16 | +0.30 |
| Incidents — and did users get paid | none to date exploits · sourced · 2026-07-15 | 25% | 10 | 9 | 1 of 16 | +0.25 |
| Audits and transparency | — | 10% | 8 | 5.5 | 1 of 16 | +0.25 |
| Cost and speed | — | 10% | 8 | 6 | 4 of 16 | +0.20 |
Measured 15 July 2026 · weights and method · decided by trust assumption, worth +1.60 points against the median
Trust assumption: 8/10
Not a label — a number. Who, exactly, can move the money if they collude or are compromised? A canonical rollup bridge answers "nobody: Ethereum itself verifies the withdrawal, and the seven-day window is the period in which fraud can be proven". An external committee answers with a threshold, and that threshold IS the rating: Ronin was 5-of-9 and four of those keys sat on one company's servers. Harmony was 2-of-5. We publish the signer count and the threshold wherever they exist, and where a project will not state them, that silence is scored as the risk it is.
Scored 8 of 10 against a category median of 4, which places it 3rd of 16 among cross-chain bridges on this criterion. At a 40% weight that is 1.60 points above the median contribution of the weighted total. The best score in the category is 10, the worst 2.
The upgrade key: 5/10
The assumption that quietly overrides the whitepaper. A bridge whose contracts can be upgraded by a small multisig HAS that multisig as its real trust model, whatever the architecture diagram says — and Nomad proves it: the bug that lost $186m was introduced BY an upgrade. So we ask who holds upgrade authority, and we score the answer rather than the design.
Scored 5 of 10 against a category median of 3, which places it 3rd of 16 among cross-chain bridges on this criterion. At a 15% weight that is 0.30 points above the median contribution of the weighted total. The best score in the category is 7, the worst 1.
Incidents — and did users get paid: 10/10
Every exploit with its root cause, and then the column nobody else publishes: were users made whole, and is that INDEPENDENTLY confirmed or merely claimed? Read the results carefully, because they do not mean what people think. Reimbursement in this category has tracked the sponsor’s balance sheet, not the bridge’s security. That makes "they paid people back last time" close to worthless as a safety signal, and we score it accordingly.
Scored 10 of 10 against a category median of 9, which places it 1st of 16 among cross-chain bridges on this criterion. At a 25% weight that is 0.25 points above the median contribution of the weighted total. The best score in the category is 10, the worst 3.
Audits and transparency: 8/10
Named firms, dated reports, published. An unnamed audit is not an audit. Undisclosed validator counts are scored as undisclosed, not assumed benign.
Scored 8 of 10 against a category median of 5.5, which places it 1st of 16 among cross-chain bridges on this criterion. At a 10% weight that is 0.25 points above the median contribution of the weighted total. The best score in the category is 8, the worst 4.
Cost and speed: 8/10
Deliberately the smallest weight, and one qualification matters more than the number: we refuse to treat the canonical bridge’s seven-day withdrawal window as a defect. It is not a delay that clever engineering removed — it is the exact period during which Ethereum can be used to prove a withdrawal fraudulent. A fast bridge is fast for one reason: it substituted somebody’s signature for that proof.
Scored 8 of 10 against a category median of 6, which places it 4th of 16 among cross-chain bridges on this criterion. At a 10% weight that is 0.20 points above the median contribution of the weighted total. The best score in the category is 9, the worst 5.
Other measurements
Its nearest neighbours in this ranking
| # | Entry | Trust assumption | How it differs |
|---|---|---|---|
| 2 | Optimism / Base canonical bridges (OP Stack) | 10 | Ahead by 2 on trust assumption. |
| 3 | Circle CCTP | 8 | Level on trust assumption; the gap is elsewhere. |
| 5 | Hop Protocol | 8 | Level on trust assumption; the gap is elsewhere. |
| 6 | Chainlink CCIP | 6 | Behind by 2 on trust assumption. |
Questions about this score
How can Across be fast if canonical withdrawals take seven days?
+
Because a relayer pays you out of its own capital immediately, then waits the seven days itself to be reimbursed through the canonical bridge. You are not skipping the security window — someone with a bond at stake is standing in it for you, and their fee is the price of that.
What happens if a relayer fails?
+
Relayers post bonds precisely so that a failure is covered, and the system is designed so that the canonical settlement underneath eventually resolves regardless of any individual relayer's fate. That is a much narrower failure mode than a committee whose collusion would drain a pot of collateral.