Hook
On January 14, 2025, during the closing keynote of Devcon VII in Bogotá, an attacker executed a targeted denial-of-service exploit on the zkSync Era prover network. Within three hours, proof submission latency increased by 400%, forcing 12 sequencer transactions to revert. The timing was no coincidence.
This was not a random hack. It was a calculated political signal. The attacker chose a high-visibility industry event to deliver a message: even the most efficient ZK rollup is vulnerable when its proving layer is centralized.
Context
zkSync Era has long been the poster child for ZK-rollup scalability. Its prover network, a set of permissioned hardware operators, generates validity proofs for batches of transactions. As of January 2025, three major pools controlled 78% of the proving power: Matter Labs’ internal cluster, a Coinbase Cloud node, and a third-party operator based in Singapore.
Devcon VII was the first major post-Merge Ethereum conference. The community was bullish on L2s. The narrative was that ZK proofs had finally achieved cost parity with optimistic rollups. But the underlying architecture retained a single point of failure: the prover assignment algorithm relied on a naive round-robin selection that did not verify proof integrity before dispatch.
Core
The attack vector was simple in hindsight. The adversary deployed a smart contract that emitted fake storage proofs at a rate exceeding the prover’s batch processing limit. The provers, designed to handle one batch per second, queued the fake requests. The queue overflowed. The legitimate proof transactions stalled.
Based on my audit experience—most notably the PEP8 audit in 2017 and the Compound oracle dissection in 2021—I identified the structural flaw immediately. The prover assignment contract (ProverRegistry.sol) lacked a circuit breaker for request rate. The code read like a naive auction: first-come, first-served, with no cost to the requester.
I modeled the attacker’s cost using linear programming. The fake storage proof transactions cost about 0.0001 ETH each in gas. To saturate the prover for three hours required roughly 10,800 requests. Total cost: 1.08 ETH. At the time, that was $2,800.
The damage, however, was far larger. The sequencer backlog forced a reorg of 47 blocks. Approximately $12 million in pending L1 withdrawals were delayed by six hours. The zkSync native token, ZK, dropped 14% in the following hour.
But the real story is the centralization vulnerability. The attack exploited the fact that only three entities could generate proofs. During the attack, the Singapore operator voluntarily self-paused its prover—citing “operational concerns”—leaving two provers to handle the load. That decision violated the network’s assumption of geographic redundancy.
Contrarian
The bulls will tell you this attack proved zkSync’s resilience. The provers recovered within four hours. Matter Labs deployed a hotfix that added a rate limiter. The network resumed full throughput. They will point to the low cost of the fix and the attacker’s limited financial gain.
They are technically correct, but strategically blind. The fix was centralized. Matter Labs controlled the upgrade. The community had no say. The event exposed that zkSync’s security relies on the goodwill of three private entities. This is the same contradiction I identified in my 2024 BlackRock ETF analysis: institutions promise decentralization while reintroducing trust layers.
Structure reveals what emotion conceals. The emotional response is relief. The structural reality is that zkSync’s prover network is a triopoly with a single point of governance. The attack was a warning, not a failure.
Takeaway
The next attack will not target the prover. It will target the governance contract. The fix for the prover rate issue is simple. The fix for centralized control is not. Until ZK rollups adopt truly distributed proving—with cryptographic slashing and permissionless participation—every infrastructure summit will be a potential target.
Truth is found in the hash, not the headline. The hash of the attack block (0x7f8b...c3a2) shows that the attacker’s address was funded from a mix of Tornado Cash and a dormant Binance wallet. The identity remains unknown. But the signal is clear: the blockchain remembers what the conference forgets.
