Across
Why it happened
Attack class: Bridge exploitOff-chain verification, not a smart contract.
What would have stopped it
- Verify every deposit event against the program's real event marker and transaction status.
- Cross-check deposits against token balance changes, not just logs.
What the score saw
Our pre-hack assessment of Across flagged cross-chain composition as its weakest area, with each chain connection a trust boundary, but it covered the on-chain contracts, not the relayer software that reads Solana deposits. The on-chain design it credited is why the loss stayed with the relayer and never reached users or liquidity providers.
Score dimensions
- Compositional riskpartly flagged
- Operational securitymissed
- Adversarial resiliencepartly flagged
An attacker sent 1,627 fake deposits on Solana that Across's main relayer, run by Risk Labs, read as real, and the relayer paid out about 4.5 million dollars of its own money on 18 other chains before Solana was switched off. No user or liquidity provider funds were lost, and every genuine transfer was completed or refunded the same day. The remaining fake requests, worth about 37 million dollars, expired unpaid. On July 28 the attacker returned 1,728.6 ETH under a 10 percent bounty deal, reported as about 3.7 million dollars, and about 500,000 dollars of attacker funds stayed trapped in the protocol.
The attacker's Solana wallet sent fake deposits that Risk Labs' relayer filled on 18 chains, paying a single recipient; the proceeds were swapped to ether, and 1,728.6 ETH came back to the Across multisig on July 28.
Attack flow
All 5 on-chain stepsThe attack in 4 steps. Step 1, Root control: Risk Labs relayer (source). Step 2, Fills: Across Protocol Exploiter 2 (attacker), about $4.5M reached this step. Step 3, Swap & bridge: Bridged and swapped to ether (bridge), about $4M reached this step. Step 4, Return: Across Hub Pool owner multisig (exit), about $3.7M reached this step.
Step 1: Root control
- SourceRisk Labs relayer581 fills on 18 chains~ $4.5M own capital0x07aE8551...e70E67
Step 2: Fills
- AttackerAcross Protocol Exploiter 2one recipient on every chain~ $4.5M received0xA6fb971F...560189
Step 3: Swap & bridge
- BridgeBridged and swapped to etherfills moved to EthereumUSDC swapped to ETH
Step 4: Return
- ExitAcross Hub Pool owner multisig1,728.6 ETH returned July 2810% bounty agreed0xB5247353...F03715
Why each step worked
- Event reader trusted look-alike dataSolana deposit events were not checked for authenticity
- 1,627 fake deposits8bkoZToaTBBtAPczgHqD4XVxWtvBkiy4crtexEtYDYSL
- No per-recipient fill limitmissing controlhundreds of fills to one address in an hour
- No balance check before fillingmissing controlno USDC ever reached the Solana pool
- Fast manual shutdownaddresses blacklisted in minutes, Solana paused within hours
Full forensic detail
Step-by-step reconstruction, root cause, counterfactuals, remediation, and disclosure timeline.
Exploit anatomy
Root cause
The root cause was off-chain verification, not a smart contract. Across relayers front their own capital to pay users on the destination chain, then get repaid later once the deposit is proven. To decide whether a Solana deposit is real, Risk Labs' relayer reads events from Solana transactions. Per the Across post-mortem, that reader accepted data aimed at the Solana pool as a genuine deposit event without checking the marker that only real events carry, so transactions that deposited nothing could pass as deposits. The attacker used that to submit 1,627 fake deposits from single-use wallets, all paying one EVM recipient, and the relayer filled 581 of them, about 4.5 million dollars, before Across cut Solana off. The on-chain design kept the damage contained: because relayers are only repaid for deposits that really happened, the loss stayed with Risk Labs' relayer, and no user or liquidity provider funds were lost. A related flaw in the same Solana event handling, where failed transactions could be read as deposits, had been disclosed and fixed in April 2026, which shows how fragile this off-chain reading layer is. Most of the money came back: the attacker returned 1,728.6 ETH on July 28 under a 10 percent bounty agreement, and about 500,000 dollars stayed trapped in the protocol.
Prevention analysis
Checking that an event is a genuine one emitted by the pool, and that the deposit actually moved tokens, would have rejected every forged request before any fill.
None of the forged deposits moved USDC into the pool. A relayer that confirms the pool balance rose before filling would have paid nothing.
Hundreds of large fills to one address from fresh wallets in an hour is a clear pattern. A cap per recipient or per hour would have stopped the payout after the first few fills; Risk Labs says it is now building this.
Similar incidents
A bridge accepted a message that only looked authentic on the Solana side, and value was paid out on another chain for a deposit that never happened. Same class: missing authenticity check on cross-chain input.
Payouts on one chain were triggered without proper verification of the deposit on the other, the same week as the Across attack.
A verification gap let anyone present an unproven message as valid, and many copycat requests drained the bridge. Same pattern of mass forged requests against one weak check.
Remediation
Timeline
Get your protocol scored across 12 dimensions, or request ongoing coverage.