Aztec Connect second exploit analysis

On June 17, 2026, three days after the first Aztec Connect drain, the deprecated project's RollupProcessor was hit again for the same $2.1 million. SlowMist founder Cos published a preliminary analysis identifying the mechanism: the contract's escape hatch accepted any rollup proof during an open window and paid out from the contract's own reserves based on the proof's public output field, without verifying that the inner transaction corresponded to a real deposit.

The escape-hatch path

The attacker constructed proof data declaring withdrawals of 1,158 ETH, 150,000 DAI and 0.46963295 renBTC from the RollupProcessor's custodial reserves. The contract called its own withdraw(publicOutput, outputOwner, assetId) function for each, paying the attacker's address from reserves it controlled. The Aztec Foundation confirmed that the deprecated product had been sunset in 2022 and that Aztec Labs retains no controls over the system. The transaction hashes are public: 0xab306cd2, 0x5c196c37 and 0x9e1d6ab7.

Why the second drain was possible

The first drain on June 14 reportedly took 909 ETH and was widely covered, but the contract was not paused and the underlying bug was not fixed. The protocol had been deprecated four years; nothing on the project side could intervene. The same RollupProcessorV3 contract was therefore live and accepting the same class of malformed proofs three days later, and a different attacker (or the same one with a different plan) used them.

Where it sits in the corpus

Aztec Connect is the only 2026 incident in the corpus in which the same deprecated protocol was drained twice in three days by similar mechanisms. The broader pattern of legacy-contract drains is shared with Raydium's deprecated AMM V3 and DxSale's locker, but those were each hit once. The Aztec case shows that the failure mode repeats against the same target if nothing about the target's code or custody changes between hits.