On 30 October 2026, a few hours after the Binance hack, an address on Arbitrum One held about 8,426 ETH and about 74.6 million ARB belonging to the exploiter. At 00:08 UTC the funds moved without the owner signing anything. The Arbitrum Security Council, a group of twelve keyholders whose emergency actions need nine signatures, used its power to upgrade the chain’s contracts on Ethereum and shifted the tokens to an address the attacker could not touch. No fraud proof was involved, and nobody waited a week.
Optimistic rollups live with that contradiction today. ETHBEAT, the risk tracker formerly called L2BEAT, logs the action on its Arbitrum One page, with 74,559,805.2 ARB and 8,425.998 ETH in a later update. The same chain where any stranger can prove a wrong result and win the bond of the cheater also has a handful of keys able to decide who owns what. Both machines are taken apart below: the challenge game that lets Ethereum check a rollup only when someone objects, and the councils that sit above it, on the three largest optimistic chains, Arbitrum One, OP Mainnet and Base, which went three different ways in 2025 and 2026.
The basics come first in our lesson on how layer 2 rollups make Ethereum cheaper: a rollup executes transactions away from Ethereum, posts their data and a summary of the result back to it, and keeps users’ funds in a bridge contract on Ethereum. Why Ethereum needs this at all is the subject of why blockchains need scaling, and our comparison of layer 1 and layer 2 already gave optimistic and zero-knowledge rollups two paragraphs. That lesson promised a closer look at how the proofs are built and how the games run, and it starts here.
Optimistic rollups explained: Ethereum checks only when someone objects
A rollup has to convince Ethereum that its new state is correct, otherwise the bridge contract cannot know who may withdraw what. There are two ways to do it. A zero-knowledge rollup attaches a mathematical proof to every batch. An optimistic rollup attaches nothing and waits. In ethereum.org’s words, optimistic rollups are called that because they “assume offchain transactions are valid”. The claim stands unless somebody shows it is wrong within a fixed window, which ethereum.org calls the challenge period.
The whole design rests on one assumption: at least one honest party is watching and will act. No majority is needed, and no committee; a single honest node is enough. If a single node runs the rollup’s software, recomputes each claimed result and finds a mismatch, it can start a dispute that Ethereum referees. If nobody watches, a dishonest proposer could, in principle, post a false state and drain the bridge once the window closes. ethereum.org lists this reliance on “at least one honest node” among the model’s drawbacks, next to the week-long wait for withdrawals and the centralized sequencers that order transactions.
The trade is worth it for one reason: proving is expensive and disputes are rare. An optimistic rollup only pays for a proof when someone contests a claim, so the normal cost of running the chain is the cost of posting data to Ethereum. The price is time. Ethereum cannot treat a claim as final until the challenge window has passed with no successful objection, and that wait is what users feel when they withdraw.
The challenge game: from a whole block to one instruction
Ethereum cannot re-execute hours of rollup activity. A single Ethereum block holds at most 60 million gas, and one transaction at most about 16.8 million since the Fusaka upgrade of December 2025, as our lesson on Ethereum gas and EIP-1559 explains. So the dispute is designed to shrink until what is left fits in one Ethereum transaction. Both modern systems, Arbitrum’s BoLD and the OP Stack’s fault proofs, follow the same three steps.
Proposers post a claim and a bond
A proposer publishes a claim on Ethereum: after these transactions, the state of the rollup has this fingerprint, a hash called the state root or output root. The claim comes with a bond that the proposer loses if the claim is proven false. On OP Mainnet the bond for a proposal is 0.08 ETH, and ETHBEAT notes that anyone can be a proposer. On Arbitrum One the stake behind an assertion is far larger, 3,600 ETH according to ETHBEAT’s Arbitrum page as of 11 November 2026, with a new assertion possible every 15 minutes.
Anyone who disagrees posts a counter-claim with its own bond. From that moment the two sides are locked in a game whose rules live in an Ethereum smart contract.
Bisection: halving the disagreement
The two parties agree on where they started and disagree on where they ended. So the contract asks them to split the work in half and say what the state was at the midpoint. If they agree at the midpoint, the error is in the second half; if not, it is in the first. Then they split again. Each round halves the disputed range, which is why the method is called bisection, and a range of billions of steps falls to one step after a few dozen rounds.
Arbitrum’s BoLD documentation describes three levels. First the parties narrow the dispute to one Arbitrum block. Inside that block they narrow it to a range of WebAssembly instructions, currently 223 steps, about 8.4 million, the so-called big step. Then they narrow that range to a single instruction. Each level needs fresh stakes: ETHBEAT lists 555 ETH at the first sub-level and 79 ETH at the second on Arbitrum One. On OP Mainnet the game is a tree of claims and counter-claims with a maximum depth of 73, and the bond required grows at each level, by a factor of about 1.095 per level according to ETHBEAT.
The bonds have a purpose. Every move costs the mover money that the honest side collects at the end, so a cheater cannot spam the game for free. ETHBEAT still flags a residual concern for Arbitrum: the gap between the stake of an attacker and the stakes needed to defend could, in theory, be used to exhaust a defender’s resources.
The one-step proof Ethereum can afford to run
When only one instruction is left, Ethereum runs it. Both sides submit what the machine looked like before that instruction, and a contract on Ethereum executes the single step and sees who told the truth. Arbitrum does this with a contract called OneStepProof, which emulates one instruction of its WebAssembly-based machine. OP Mainnet uses Cannon, a MIPS virtual machine: the rollup’s state-transition program is compiled to MIPS so that any single step can be replayed on Ethereum.
The loser forfeits its bonds. The false claim is removed, and the honest claim, or a new correct one, carries on toward confirmation. Nothing about the rollup’s past transactions changes; only the wrong summary of them is thrown away.
Arbitrum’s BoLD: why anyone can now defend the chain
Arbitrum ran fraud proofs long before BoLD, but only allow-listed validators could take part. The old protocol also had a weakness known as a delay attack. Disputes were played one against one, in sequence, so an attacker with enough money could open challenge after challenge and stall confirmation for a long time even without ever winning.
BoLD, short for Bounded Liquidity Delay, changes the format. Bonds are attached to claims rather than to people, so any number of honest, even anonymous, parties can defend the same correct claim, and disputes can run in parallel. The Arbitrum docs put the worst case at “two challenge periods, plus a two-day grace period” reserved for the Security Council, plus a small allowance for computation. ETHBEAT computes the maximum time to confirm a disputed claim on Arbitrum One at 12 days 17 hours, if both sides play as slowly as the rules allow.
The Arbitrum documentation dates BoLD’s activation on Arbitrum One and Arbitrum Nova to 09:00 Eastern time on 12 February 2025, after a DAO vote. On Arbitrum One, validation is now permissionless. Nova kept its permissioned validators, as The Block reported at the launch. The challenge period on Arbitrum One is 6 days 8 hours on ETHBEAT, about 6.4 days, which most bridges and wallets round to “a week”.
OP Mainnet’s fault proofs: Cannon, a pause in August 2024, and Kona
OP Mainnet’s history is less tidy. Its first fraud-proof system was removed in November 2021, when the chain moved to OVM 2.0, and for about two and a half years OP Mainnet ran with no working proof at all: withdrawals relied on a permissioned proposer. According to the Optimism documentation, “On June 10, 2024, Fault Proofs were officially added to the OP Stack and were activated on OP Mainnet.” From then on, anyone could propose an output root and anyone could challenge one.
The system did not run uninterrupted. ETHBEAT’s OP Mainnet page records that fault proofs were disabled on 16 August 2024 after a bug was found, with the chain falling back to permissioned proposals. The fix came with the Granite upgrade, live on 28 August 2024, and ETHBEAT counts 26 days in all before the proofs were back.
The OP design adds a second safety layer. After a dispute game resolves, an extra delay called the airgap window starts before withdrawals can use the result, and during that window a Guardian role can reject the root. On ETHBEAT’s figures, the challenge clock runs 3 days 12 hours and the execution delay another 3 days 12 hours: 7 days in total. The Optimism docs present the modular design as groundwork for several proof systems running side by side, so that a bug in one is not fatal. ETHBEAT shows that a newer “super” dispute game, built for interoperability between OP Stack chains, already uses a program from Kona, a second implementation of the fault proof program. A full multi-proof setup on OP Mainnet remains a plan, with no date.
OP Mainnet is run by the Optimism Collective, whose project page on Coindaro is Optimism. Its Security Council has thirteen members and acts with ten signatures; upgrades also go through a two-of-two multisig shared by the Optimism Foundation and the Council, with a 7-day veto window for normal changes.
Base in 2026: an optimistic window with ZK and TEE proofs
Is Base an optimistic rollup? It started as one, built by Coinbase on the OP Stack and launched in 2023, and it grew into the largest member of Optimism’s Superchain. In 2026 it moved on. On 18 February 2026 Base’s engineering team announced that it would replace its OP Stack dependencies with a single Base-operated stack in one repository, base/base, built on open-source parts such as the staking means depositing a proof of stake coin with a protocol that stakes it for you and issues a transferable token (such as Lido stETH or rETH) representing the deposit plus rewards, which you can use or sell while the coins stay staked.">Reth client, while remaining compatible with the OP Stack specification. “The protocol spec and codebase should be understandable by a single developer,” the post said. The same post said an additional independent signer would join Base’s Security Council to replace Optimism’s role.
Money moved too. DL News reported on 19 February 2026 that Coinbase would stop sharing 15% of Base’s revenue with the Optimism Collective. Base had contributed 8,387 ETH, about 41% of the Collective’s lifetime revenue, according to the same report, and the OP token lost about a quarter of its value within two days of the news.
The bigger change for users came with Azul, which the Base upgrades page dates to 28 May 2026 (ETHBEAT gives 26 May). Azul replaced the interactive fault-proof game with what Base calls a multiproof. A proposal now covers 6,000 Base blocks and carries a bond of 0.05 ETH. It is backed by one or both of two kinds of evidence: an attestation from a trusted execution environment, a sealed AWS Nitro enclave that vouches for the computation, and a zero-knowledge proof generated with Succinct’s SP1. ETHBEAT’s Base page explains the clock: when both the TEE and the ZK proof are submitted, a proposal resolves after 1 day; with a single proof it takes 5 days. That second figure was 7 days when Azul went live, until the Beryl upgrade of 25 June 2026 shortened the delay, per Base’s upgrade schedule. Anyone with a valid ZK proof that a root is wrong can challenge it and take the proposer’s bond.
So Base keeps an optimistic window, which is why ETHBEAT still files it under optimistic rollups, but it no longer plays the bisection game. Disputes are settled in a single round by a proof, as on a ZK rollup. Our glossary entry on zero-knowledge proofs explains what such a proof shows. The TEE side brings its own trust assumption: the list of approved TEE provers is managed by the Base Coordinator Multisig, a 3-of-6 Coinbase multisig, without a vote of the Security Council, according to ETHBEAT.
Base has kept changing at speed. The Cobalt upgrade went live on 30 September 2026 according to Base’s documentation, and Denim cut block time from 2 seconds to 200 milliseconds on 5 November 2026, retiring the Flashblocks system, as planned in our report on the Denim schedule. As of 11 November 2026 Base secures about $13.8 billion on ETHBEAT, the most of any layer 2.
Why your withdrawal waits about a week, or less on Base
A deposit from Ethereum to a rollup takes minutes. A withdrawal through the official bridge takes days, because the bridge contract on Ethereum may only pay out against a state root that has survived its challenge window. On Arbitrum One that is about 6.4 days, on OP Mainnet 7 days, on Base 1 to 5 days depending on how many proofs back the proposal. The step-by-step process, from starting the withdrawal on the layer 2 to proving and finalising it on Ethereum, is in our guide to bridges, liquidity and intents.
Most people never wait. Third-party bridges and liquidity providers pay the user on Ethereum within minutes, take over the user’s claim on the rollup, and wait out the window themselves for a fee. That moves the risk rather than removing it: the user now trusts the fast bridge’s contracts and operators for the few minutes of the transfer, and the provider carries the week on its balance sheet. The official bridge is slower, but it only asks you to trust the rollup itself. A dedicated guide on how to bridge to a layer 2 will come later in this path.
When the sequencer censors or stops: forced inclusion
On all three chains, a single operator runs the sequencer, the machine that orders transactions and produces blocks; ETHBEAT lists the sequencer as centralized on each. A sequencer cannot steal funds or forge a state that survives a challenge, but it can ignore you or go offline. Optimistic rollups answer with a back door on Ethereum.
On Arbitrum, a user can send a transaction to the delayed inbox, a contract on Ethereum. If the sequencer does not include it, anyone can force it into the chain after a delay. The Arbitrum docs set that delay at the lesser of 7,200 Ethereum blocks, about 24 hours, and a delay buffer that ranges from 30 minutes to 48 hours and shrinks when the sequencer is caught being slow. On OP Mainnet and Base, the path is a deposit through the OptimismPortal contract on Ethereum, which the chain must include within 12 hours, the sequencing window of 3,600 Ethereum blocks, according to ETHBEAT.
Forced inclusion is a slow emergency exit, not a daily tool. It also says nothing about who orders transactions in normal times. Arbitrum adopted Timeboost, an auction for a short head start in ordering, in 2025; ETHBEAT recorded in an update of 28 September 2026 that it has been replaced by priority gas auctions through a Constitutional AIP. The economics of ordering and front-running belong to our guide to MEV.
Where the data goes: blobs, calldata and Arbitrum Nova’s committee
A fraud proof works only if the challenger can rebuild the rollup’s state, so the transaction data must be public. Rollups first posted it as ordinary transaction data on Ethereum, called calldata. Since the Dencun upgrade of 13 March 2024 they can use blobs, a cheaper space that Ethereum keeps for about 18 days, long enough to cover any challenge window. OP Mainnet switched with its Ecotone upgrade on 14 March 2024, per ETHBEAT. How blobs are priced belongs to the gas lesson linked above, and a separate lesson on blobs and EIP-4844 is planned.
Posting has to be regular. ETHBEAT logged a gap on Arbitrum One on 29 October 2026, from 20:31 to 20:48 UTC: 17 minutes 24 seconds with no transaction data sent to Ethereum, against a usual interval of about 1 minute 50 seconds. Transactions kept being confirmed by the sequencer during that time, but they were not yet anchored on Ethereum.
Arbitrum Nova takes a different route. It uses AnyTrust: instead of posting all data to Ethereum, it hands it to a Data Availability Committee of seven parties, which sign that they hold it, according to the Arbitrum FAQ. If too few members sign, Nova falls back to posting the data on Ethereum like a normal rollup. The saving comes with an extra trust assumption, and ETHBEAT rates Nova Stage 0. The chain is close to idle: growthepie counted 791 transactions on 10 November 2026.
Security Councils: the keys above the fraud proofs
A fraud proof protects users against a false state. It does not protect them against an upgrade. Each of the three chains can replace its own contracts on Ethereum, and each has given that power to a council that can act at once in an emergency. Arbitrum’s Security Council has twelve members in two cohorts of six and needs nine signatures; normal, non-emergency upgrades go through timelocks of 8 days on the rollup and 3 days on Ethereum, per ETHBEAT. OP Mainnet’s council acts with ten of thirteen. On Base, upgrades need both a 3-of-6 Coordinator Multisig run by Coinbase and an 8-of-11 Security Council, and ETHBEAT notes that upgrades take effect as soon as both sign, so users have no exit window before a change.
These powers are not theoretical. On 21 April 2026, at 23:26 Eastern time, Arbitrum’s council moved 30,765.67 ETH held on Arbitrum One by the attacker of KelpDAO. It upgraded the Inbox contract on Ethereum to a temporary version, added a function that acts as the exploiter’s address, sent the funds to an address ending in 0DA0, and restored the original code, in what the council called an atomic action. “To release the funds, a subsequent action must be taken by Arbitrum Governance,” the council wrote. Six months later, on 30 October 2026, it froze the Binance exploiter’s ETH and ARB described at the top of this page; the hack itself is told in our report on the Binance hack. On 2 October 2026 the same council paused new activations of Stylus, Arbitrum’s system for contracts written in Rust, C and C++, citing AI-assisted attacks on hand-crafted WebAssembly programs that could slow the chain; existing programs kept running. A plan to reopen activations is covered in our report on the Stylus pause.
The April freeze split the industry, as FXStreet reported the same day. Ledger’s chief technology officer Charles Guillemet saw a known feature of the design used in the open: “This is not a secret or an abuse.” Curve founder Michael Egorov accepted that the action may have been justified but warned that users could start to doubt how neutral these networks are. For a user, the practical point is narrower. On these chains a balance is only as safe as the weaker of two things: the fraud proof and the council.
This is why all three chains sit at Stage 1 on ETHBEAT’s scale and none at Stage 2. Stage 1 requires a working proof system that only a large majority of a Security Council can override. Stage 2 requires that upgrades unrelated to provable bugs give users at least 30 days to leave, and that the council can only step in over bugs proven on-chain. On ETHBEAT, all three fail those two tests. A later lesson on layer 2 risks and ETHBEAT’s stages will go through the full checklist.
Arbitrum, OP Mainnet and Base compared
The figures below come from ETHBEAT’s project pages as of 11 November 2026 and from growthepie’s public data for the day of 10 November 2026. Activity counts on rollups include bots and system transactions, so OP Mainnet’s high count next to tiny fees should be read as transactions recorded, not as people.
| As of 11 Nov 2026 | Arbitrum One | OP Mainnet | Base |
|---|---|---|---|
| ETHBEAT stage | Stage 1 | Stage 1 | Stage 1 |
| How claims are checked | BoLD interactive fraud proofs | Cannon fault proofs | TEE attestation and SP1 ZK proof, single round |
| Normal withdrawal wait | About 6.4 days (6 d 8 h) | 7 days (3.5 + 3.5) | 1 day with both proofs, 5 days with one |
| Forced inclusion | Up to about 24 hours | Up to 12 hours | Up to 12 hours |
| Emergency council | 9 of 12 | 10 of 13 | 8 of 11, plus Coinbase 3 of 6 |
| Value secured (ETHBEAT) | $9.03bn | $1.42bn | $13.82bn |
| Transactions, 10 Nov 2026 | 1,289,367 | 1,859,879 | 8,468,276 |
| Daily active addresses, 10 Nov 2026 | 107,461 | 393,928 | 348,168 |
| Median transaction cost, 10 Nov 2026 | $0.0048 | $0.000025 | $0.0018 |
| Fees paid by users, 10 Nov 2026 | $21,143 | $2,826 | $130,798 |
For scale, Ethereum itself processed 2,064,096 transactions on 10 November 2026 at a median cost of about $0.03, according to the same growthepie data, and collected about $349,656 in fees. Base alone handled four times as many transactions as Ethereum that day for a median cost about sixteen times lower. More on the team and the DAO behind Arbitrum is on our Arbitrum project page.
Optimistic vs ZK rollups: finality, withdrawals and proof cost
The question “optimistic or ZK?” comes down to when the proof happens. An optimistic rollup proves only when challenged, and so must wait. A ZK rollup proves every batch, and so can be final as soon as the proof is verified on Ethereum. Each choice moves a cost somewhere else.
| Optimistic rollup | ZK rollup | |
|---|---|---|
| What is posted to Ethereum | Data plus a claimed state root | Data plus a claimed state root and a validity proof |
| When Ethereum accepts the state | After the challenge window, if no one wins a dispute | Once the proof is verified |
| Withdrawal through the official bridge | About a week on Arbitrum One and OP Mainnet | Hours, depending on how often proofs are posted |
| Proof cost | Paid only when a claim is disputed | Paid for every batch |
| EVM compatibility | Close to native from the start | Historically harder; zkEVMs have narrowed the gap |
| Security assumption | At least one honest, active challenger | A correct proving system and verifier contract |
| Hybrid case | Base since May 2026: optimistic window, single-round TEE or ZK proofs |
Neither model removes the councils described above, and neither removes the centralized sequencer. The differences are in exits and costs. ZK rollups such as zkSync Era, Starknet, Linea and Scroll, with their circuits and provers, get a lesson of their own on zero-knowledge rollups later in this path.
Outages and interventions that really happened
The big optimistic chains have run for years now, long enough to have a record. None of the events below led to a successful fraud against a bridge, but several show where the trust really sits. Arbitrum’s longest stop so far came on 15 December 2023, when a wave of inscriptions, small data messages copied from Bitcoin’s Ordinals, stalled the sequencer for about an hour and a half, according to The Block.
| Date | Chain | What happened | Source |
|---|---|---|---|
| 15 Dec 2023 | Arbitrum One | Sequencer stalled during a surge of inscriptions; outage of about an hour and a half | The Block |
| 16 Aug 2024 | OP Mainnet | Fault proofs disabled after a bug for 26 days; fix shipped with Granite (28 Aug 2024) | ETHBEAT |
| 5 Aug 2025 | Base | Block production halted 33 minutes after a failed sequencer handover | Base postmortem via The Block |
| 21 Apr 2026 | Arbitrum One | Council moves 30,765.67 ETH of the KelpDAO exploiter | Arbitrum forum |
| 2 Oct 2026 | Arbitrum One | New Stylus activations paused by the council | ETHBEAT |
| 29 Oct 2026 | Arbitrum One | 17 min 24 s without data posted to Ethereum | ETHBEAT |
| 30 Oct 2026 | Arbitrum One | Council freezes about 8,426 ETH and 74.6M ARB of the Binance exploiter | ETHBEAT, Coindaro |
Base’s 2025 halt is a good example of sequencer risk in practice. According to The Block’s account of Base’s postmortem, the active sequencer fell behind under heavy on-chain activity, and the automatic handover went to a new sequencer that was still being set up and could not pass control on again. Blocks stopped. Engineers paused the handover system, called Conductor, and moved block production to a healthy sequencer by hand. For 33 minutes, deposits and withdrawals waited along with everything else.
ARB and OP: governance tokens, not a claim on the chain
ARB and OP are voting tokens. Neither entitles its holder to the sequencer’s revenue or to the funds in the bridge. ARB was created with an initial supply of 10 billion tokens and a cap on new issuance of 2% per year, according to the Arbitrum Foundation’s documentation; holders vote in the Arbitrum DAO, which elects the Security Council and approves upgrades of Arbitrum One and Nova. OP is the governance token of the Optimism Collective, where a proposal from October 2026 would end Security Council elections, as our report on Optimism’s governance proposal explains. ETHBEAT counts OP tokens for about 15.8% of the value secured on OP Mainnet. Base has no token.
The Base split shows how loosely a token is tied to the technology it governs. Base ran on the OP Stack, paid revenue to the Collective and helped make it the reference design for optimistic rollups, but OP holders had no say over Base’s departure. Prices of both tokens move for many reasons; nothing in this lesson is a view on them. What a holder actually gets is a vote on the councils and timelocks described above. Arbitrum’s DAO, for instance, will decide what happens to the 30,765.67 ETH frozen in April.
A later lesson in this path turns to zero-knowledge rollups, the model that proves every batch instead of waiting for a challenge.
