A fraud proof is a cryptographic demonstration submitted to Ethereum L1 proving that an optimistic rollup sequencer published an incorrect state transition. If successful, the fraudulent batch is reverted and the sequencer is penalized. Fraud proofs are the security backbone of optimistic rollups: they make cheating unprofitable even though the system initially trusts submissions.
What Is a Fraud Proof?
3 min read
The short version
A fraud proof is like calling out a referee who made a bad call, and having instant replay prove you right. The game (L1 settlement) initially trusts the referee (sequencer), but anyone in the stands can challenge. If the replay (re-executing the disputed step on L1) shows the call was wrong, the referee is penalized and the play is overturned.
How It Works
How interactive fraud proofs work (Arbitrum model): (1) A challenger stakes a bond and claims "this output root is wrong." (2) The sequencer and challenger engage in a bisection game on L1: they repeatedly narrow the dispute by splitting the contested execution range in half, each party asserting their version of the state at each midpoint. (3) After O(log n) rounds (about 20-30 on-chain interactions over ~1 week), the dispute narrows to a single computational step. (4) That single step is re-executed on L1 in the EVM. The L1 result definitively shows who was right. (5) The wrong party loses their bond. If the sequencer was wrong, the batch is reverted. How single-step fraud proofs work (Optimism/Cannon): the entire L2 execution is compiled to a format the L1 can execute one step at a time. The challenger identifies the exact step where the error occurred and submits it for L1 verification in one transaction. Security assumption: at least one honest watcher must exist who will submit fraud proofs when needed. If all watchers are offline or colluding with the sequencer, fraud could go unchallenged.
An Arbitrum bisection game catching fraud
A sequencer posts output root claiming Alice's balance is 100 ETH after a batch. A watcher's local computation shows Alice should have 95 ETH (the sequencer inflated it by 5 ETH). The watcher challenges, staking a 1 ETH bond. Round 1: the batch contained 10,000 instructions. Challenger says "the error is in the first 5,000." Sequencer says "those are correct." They compare state at instruction 5,000. Disagreement found. Round 2: narrow to instructions 2,500-5,000. Repeat. After ~14 rounds (log2 of 10,000 ≈ 14), they reach one instruction. L1 executes that instruction and finds the sequencer's version is incorrect (they counted 5 ETH that should not exist). Sequencer loses their bond. The batch is reverted to the correct state. Alice has 95 ETH as she should.
What People Get Wrong
Fraud proofs fire automatically
Someone must actively monitor the rollup and submit proofs. If no one is watching (or all watchers are compromised), fraud could go unchallenged and become final after 7 days. This is why having multiple independent watchers is critical for optimistic rollup security.
Fraud proofs are submitted constantly
In practice, fraud proofs are extremely rare because sequencers have economic incentives to be honest (they lose their bond if caught). The mechanism exists as a deterrent. The fact that fraud proofs CAN be submitted is what prevents fraud, not the frequency of actual submissions.
The 7-day period could be shortened easily
The length is a security parameter. Shorter periods reduce the time watchers have to detect fraud, submit proofs, and complete the bisection game. The 7 days accounts for potential L1 congestion, watcher response time, and the interactive game duration. Reducing it weakens security.
Keep Reading
Sources & Further Reading
- Arbitrum Dispute Protocol
Technical specification of the interactive bisection fraud proof game
Questions People Also Ask
- Are fraud proofs live on Arbitrum and Optimism?
- Arbitrum has had fraud proofs live since launch (initially with a permissioned set of challengers, expanding toward permissionless). Optimism's Cannon proof system launched on mainnet in 2024. Both are functional but with varying degrees of decentralization in who can submit challenges.
- What happens to users during a fraud proof dispute?
- Nothing immediately. Transactions on the L2 continue normally. Only the disputed batch is at risk of being reverted. If your transaction was in that batch and fraud is proven, your transaction gets re-included in a corrected batch. In practice, users rarely notice disputes occurring.
- Could a well-funded attacker keep challenging valid batches?
- Frivolous challenges cost the challenger their bond (which they lose when the bisection game proves the batch was correct). Spam challenging is economically irrational because you lose your bond every time you challenge honest batches. The bond requirement makes this attack expensive and self-limiting.