Across Went 5 Clean Years Without an Attack. The First One Hit Its Safety Layer

What Happened

Across Protocol, the intent-based cross-chain bridge, confirmed a security incident on its Solana deployment at roughly 5:30 UTC on Friday, July 17. It is the first attack the protocol has publicly disclosed since launching in 2021, a run of roughly 5 years and more than $34 billion in bridged volume without a reported exploit.

The damage landed on the relayer operated by Risk Labs, the foundation that supports Across and 1 participant in the protocol's otherwise permissionless relayer network. The team says user funds are safe, every in-flight bridge transaction completed, and no user action is required.

As containment, deposits from Solana were disabled while routes on Ethereum, Base, and the protocol's other Ethereum Virtual Machine (EVM) chains continued operating.

Across is coordinating the investigation with SEAL 911, the emergency-response group of whitehat researchers, auditors, and incident responders that protocols call when funds start moving. The team published 3 addresses tied to the attacker, 1 on Solana and 2 on EVM chains, an early signal that the stolen value is already hopping networks.

As of this writing, 2 load-bearing facts remain undisclosed: how much the Risk Labs relayer lost, and how the attacker did it. The location is telling on its own. Solana is the newest and least battle-tested leg of Across: it became the protocol's first non-EVM chain in July 2025 with the V4 upgrade, which meant reimplementing the bridge's trust assumptions on an execution model that behaves very differently from the EVM.

The intent bridge's pitch is that user funds are never at risk because relayers absorb that risk for a fee. On Friday the risk found the absorber. Containment worked; the lesson is where the risk now lives.

The Shock Absorber Took the Hit

To see why this incident matters beyond its (apparently contained) size, look at what a relayer is. When a user bridges through Across, they sign an intent: a declaration of what they want delivered on the destination chain. Independent relayers compete to fill that intent, fronting the destination funds out of their own capital within seconds, long before the source-chain deposit reaches finality.

The protocol reimburses relayers on a settlement cycle that runs roughly every 90 minutes. The user gets speed precisely because the relayer absorbs the risk that the deposit they saw was real, final, and worth what it claimed. That risk transfer is the product. It is why intent bridges marketed themselves as the answer to the pool-drain bridge disasters of 2022: no giant honeypot of user liquidity sitting in a contract waiting to be emptied.

The fair counterpoint, and Across deserves it, is that the model worked. Users were made whole by construction, not by a post-hoc bailout. A relayer eating a loss is a categorically better failure mode than 9 figures of user deposits vanishing. But containment is not immunity; it is relocation.

The exploit risk did not disappear, it moved onto relayer balance sheets, and Friday showed that attackers have followed it there. It concentrated, too: the relayer that took the hit is run by the same foundation that stewards the protocol, which makes the line between 'protocol loss' and 'operator loss' thinner than the reassuring framing suggests.

And relayer losses do reach users eventually, just through slower channels: wider fees to price the newly demonstrated risk, thinner fill liquidity on riskier routes, and paused corridors, which Solana depositors are experiencing right now.

The April Warning Shot

The attack vector has not been disclosed, and any claim about the mechanism would be speculation. What is documented is that this exact surface had a near miss 3 months ago. In April 2026, security firm Asymmetric Research disclosed an event-spoofing vulnerability in Across's Solana integration.

The short version: relayers decide to front funds based on deposit events they observe on the source chain, and on Solana, events are often reconstructed from transaction traces in ways that differ from the EVM, including data emitted by failed transactions. The researchers showed an attacker could fabricate a deposit event convincing enough that a relayer would fill an order backed by no real deposit.

Across patched the bug promptly and no funds were lost. Whether Friday's attack is related to that class of flaw is not yet known, but the disclosure established the blueprint: do not attack the bridge contracts, attack the signal the relayer trusts.

That framing is what earns this incident a place in this series. The Trust Inversion pattern is a protective mechanism becoming the attack surface, and the recent record is consistent: an oracle verifier on Hedera that accepted a zeroed signature and waved through a $9 million drain, a privacy protocol whose proof verification verified nothing and lost $820,000, a rollup halted after forged proofs passed inspection.

In each case the component that existed to say 'this is safe' is the component that failed. A relayer is that component for an intent bridge: its capital and its event-parsing logic are the safety guarantee users rely on without seeing. The early evidence suggests attackers have fully internalized this, and the verification layer, not the vault, is where the profitable bugs now live.

What This Means for You

If you bridged through Across recently, current disclosures say you need to do nothing: user funds were not touched and in-flight transfers completed. The durable lesson is about posture. Every bridge hop inserts trusted machinery between you and your money: relayers, oracles, event parsers, settlement contracts.

CertiK's half-year numbers underline where that leads; wallet and infrastructure compromise was the costliest attack category of H1 2026, at more than $444 million across just 33 incidents, dwarfing losses from bugs in audited contract code. Treat bridges as transit, never as storage: complete the transfer promptly, check the protocol's official status channel before depositing, and do not leave value parked mid-route.

The sovereignty read is the same one this series keeps arriving at. Bitcoin held in self-custody, with private keys on a hardware wallet and multi-sig for larger holdings, has no relayer, no oracle, and no event parser in the loop; there is simply less trusted machinery to invert.

Cross-chain activity has its uses, but it is a custodial-risk decision every time you make it, even when no custodian is named on the label. Size your exposure to any single bridge or chain leg accordingly, and if a route you rely on has just disclosed an incident, wait for the post-mortem before resuming, because the first 48 hours of 'user funds are safe' statements are written before anyone fully knows why the attack worked.

What to Watch

3 things will define how serious this becomes. First, the post-mortem: the loss figure, the vector, and specifically whether it descends from the event-spoofing class Asymmetric Research documented in April, which would mean the patch closed 1 door and left the hallway. Second, when and how Solana deposits are re-enabled, and whether relayers demand structural changes (or higher fees) to keep fronting capital on non-EVM routes.

Third, the rest of the intent-bridge field: if the profitable target is now the solver and relayer layer rather than the bridge contract, competing protocols with the same architecture should be assumed to be under the same scrutiny, and the next disclosure will tell us whether this was an Across bug or an architecture bug.

Tracked entities this report names, with total coverage across the archive. Follow one into the Actor Network to see every report it appears in.

Other reports that name the same tracked actors — a different thread than “related,” which follows wording. These are linked by who appears in them.