Skip to main content

This site is for educational purposes only. Nothing here constitutes financial advice.

Topic 181 of 189

Rollup Security Stages & Escape Hatches

Rollup security stages rate how much users must trust a rollup's operators and council, and escape hatches are the exits that still work without them.

Beginner
8 min readUpdated October 2026Block Clarity Hub Editorial Team

The basics

Who Really Controls Your Funds on a Rollup?

A rollup is a separate chain that runs transactions away from Ethereum and posts the transaction data back to it, while the money users deposit sits in smart contracts on Ethereum. One operator, the sequencer, receives transactions, orders them and publishes them. That is fast and cheap, but 'built on Ethereum' does not say what happens if the operator misbehaves, goes offline or is pressed to censor you. Three questions cover most of it. Can the operator block your withdrawal? Can a small group change the rules of the contracts holding the funds? Can you push a transaction in through Ethereum if the operator ignores it?

The Stage 0, 1 and 2 Ladder

In November 2022 Vitalik Buterin proposed sorting rollups into three stages by how heavily they lean on 'training wheels', meaning a group of people who can override the code if it turns out to be buggy. The rating site L2BEAT later turned the idea into a published framework. Stage 0 means the people running the project remain in full control, Stage 2 means the code is in charge and humans may step in only for provable bugs, and Stage 1 sits between the two. As of October 2026, the rollups tab of L2BEAT's summary page listed 22 projects: 4 at Stage 2, 6 at Stage 1 and 12 at Stage 0.

A Stage Is Not a Safety Grade

A stage describes who holds power, not whether the code is free of bugs. L2BEAT's own explainer warns that a team can easily deploy an unsafe project that still reaches Stage 2, for example because the proof system is experimental or the contracts are unaudited. A higher stage means fewer people can overrule the code, which is useful, but it is one input rather than a verdict.

  • Stage 0: the operators can still overrule or freeze the system, so users rely on their good conduct.
  • Stage 1: proofs decide which state is valid and a security council can step in, so the question is how hard that council is to corrupt.
  • Stage 2: proofs are open to anyone, upgrades give users long warning, and the council acts only on provable on-chain bugs.
  • Escape hatch: a route through Ethereum that lets you force a transaction in without the operator.

Key Takeaways

  • A rollup operator orders transactions, but deposited funds sit in contracts on Ethereum.
  • Stages 0, 1 and 2 grade how much power humans keep over the code, from full control to almost none.
  • A stage measures decentralisation of control, not the absence of bugs.
  • An escape hatch is the exit that works even when the operator will not cooperate.

Going deeper

Forced Inclusion: Getting a Transaction In Without the Sequencer

The common escape hatch works like this: a user sends a message to a contract on Ethereum, and if the sequencer has not included it within a time limit, the rollup's rules treat it as part of the chain anyway. ethereum.org describes the same idea for optimistic rollups, where users may submit a transaction directly to the on-chain rollup contract once it has been delayed past the maximum time allowed. In Arbitrum, deposits and retryable tickets enter through the parent chain's delayed inbox, and the Sequencer Inbox has a forceInclusion method. As of October 2026, Arbitrum's documentation gives the force-inclusion window as the lesser of a delay buffer and a constant of 7200 blocks, which it equates to 24 hours.

The Optimism Version

OP Stack chains use a different door. Optimism's documentation says users can always bypass the sequencer by calling the depositTransaction function on the OptimismPortal contract on Ethereum. As of October 2026, it states that such transactions are processed in the order received and within the sequencer window, set to 12 hours by default but able to differ from chain to chain. If the sequencer is down or fails to publish in time, the chain automatically reorganises to include them. The same page separates a downtime outage, where the sequencer cannot process anything, from a submission outage, where it processes transactions but cannot publish them to Ethereum. It also admits that once a sequencer returns, putting portal transactions first is typical but not a rule.

Reading the Council and the Exit Window

Escape hatches protect against a censoring operator; the council and exit window protect against changed rules. The Security Council is a multisig able to act in emergencies. As of October 2026, L2BEAT's specification, whose latest changelog entry is dated 30 April 2026, asks for at least 8 participants and a 75% threshold, counts addresses controlled by one entity as a single signer, and wants identities disclosed. Upgrades started by anyone outside the council are allowed at Stage 1 if they give an exit window of at least 5 days, meaning the time users get to leave before an unwanted upgrade takes effect. Whereas those outside actors must give notice, L2BEAT's summary page states that the council may upgrade a Stage 1 rollup instantly.

Why the Council Threshold Matters

Take a council that needs 10 of 13 signatures. Four members are then enough to stop it acting, which L2BEAT's forum calls a quorum-blocking minority, and an L2BEAT post from January 2025 used that fact to show how bribing four members could stop a council from reversing a bad state root in one OP Mainnet scenario. A higher threshold is not automatically safer: it means fewer members need to be bribed to block the council. The Stage 1 test therefore looks at who can stop users leaving, not only at who can upgrade.

Key Takeaways

  • Forced inclusion routes a transaction through Ethereum when the sequencer ignores it, after a time limit.
  • Arbitrum's force-inclusion window is described as 24 hours, while OP Stack's sequencer window defaults to 12 hours (documentation read 27 October 2026).
  • Stage 1 allows an instant council upgrade, so the council's threshold and diversity are the real security assumption.
  • The exit window is the time users have to leave before an unwanted upgrade takes effect.

Under the hood

The Exact Requirements and Where They Came From

The 2022 proposal on Ethereum Magicians required, at Stage 0, a rollup full node that anyone could run and machinery letting users withdraw without the operator's cooperation, so that a false state root was the operator's only tool. As of October 2026, L2BEAT's source-of-truth post on its forum lists six Stage 0 questions: the project calls itself a rollup, state roots are posted on Ethereum, data availability is on Ethereum, software able to rebuild the state is source-available, a proper proof system exists, and at least 5 external actors can submit a fraud proof. The post's changelog records that the 5-challenger rule moved from Stage 1 to Stage 0 on 20 June 2025.

Stage 1 and Stage 2 in Detail

As of October 2026, the Stage 1 test centres on one property: other than through bugs, the only way to block an L2 to L1 message indefinitely, or to push an invalid one through, is to compromise at least 75% of the Security Council. If anyone with enough resources can propose, L2BEAT assumes at least one live proposer and does not assume it is non-censoring. Stage 2 adds permissionless fraud proofs, a 30-day exit window for unwanted upgrades including DAO upgrades, and a council confined to errors detectable on chain, such as two valid but contradictory proofs for the same batches. The 2022 proposal asked for 6-of-8 and a seven-day upgrade delay at Stage 1; the present text keeps the 8-member, 75% floor but adds rules such as counting one entity's addresses once.

Challenge Periods and Proof Types

An optimistic rollup is safe only if someone can post a fraud proof on Ethereum, and a censored proof could let a bad withdrawal through. That is why challenge periods exist. L2BEAT's forum post of 30 April 2026 cut the Stage 1 minimum challenge period from 7 days to 5, reasoning that, with strong 51% censorship attacks set aside, one day covers economic censorship games and four days cover network delays, while the 7-day bound for Stage 2 stays. For a validity-proof (zk) rollup, L2BEAT's Stage 0 text says the proof system is required to accept a state root, so the exposure shifts from a challenge window to the correctness of the proof system.

What the Ladder Leaves Out

L2BEAT states that a Stage 2 project can still be unsafe, and cites Vitalik Buterin's 2025 analysis of when Stage 1 and Stage 2 make sense, which weighs the chance of a bug against the chance that permissioned actors are compromised. The framework also asks whether data availability is on Ethereum, so a project whose data sits with a committee or another network fails that Stage 0 question. L2BEAT's own pages also differ on one figure. On 27 October 2026 its explainer page, last updated on 23 July 2025, and the glossary in its summary page still gave 7 days for the Stage 1 exit window, while the forum specification, which calls itself the source of truth, gave 5, and this page follows the specification.

Worked Example: Leaving After the Sequencer Stops

Suppose a sequencer stops and a user on an optimistic rollup wants out. First the withdrawal request is forced in through Ethereum and waits out the inclusion window, which as of October 2026 the documentation caps at 24 hours for Arbitrum and sets at 12 hours by default for OP Stack. Then the usual challenge period must pass before funds leave the bridge. This page does not cover the equivalent exit steps on validity-proof rollups.

Key Takeaways

  • L2BEAT's source-of-truth post lists six Stage 0 questions, one core Stage 1 property and three Stage 2 tests (read 27 October 2026).
  • The Stage 1 minimum challenge period fell from 7 to 5 days on 30 April 2026; Stage 2 keeps 7 days.
  • Stage 2 requires a 30-day exit window and limits the council to errors detectable on chain.
  • Validity-proof and optimistic rollups differ in their exit mechanics: one relies on proofs to accept state, the other on a challenge window.
  • A stage rates control of the system, not the quality of its code or audits.

Related Content

Prediction Markets Deep Dive

How Polymarket, Kalshi, and Augur-style prediction markets work — the mechanism design, liquidity structure, regulatory divide, election-cycle dynamics, and where these markets are and aren't reliable signals.

Liquid Restaking Tokens (LRTs)

How LRTs (ether.fi eETH, Renzo ezETH, Puffer pufETH, Kelp rsETH, and others) build on Lido-style liquid staking and EigenLayer restaking, why they emerged, their yield economics, and the specific risks that don't apply to plain liquid staking.

PBS and MEV-Boost

How Proposer-Builder Separation (PBS) and MEV-Boost restructured Ethereum's block-production supply chain, who the builders and relays are, why it matters for censorship resistance, and what enshrined PBS would change.

Merkle Trees & Merkle Proofs

A Merkle tree condenses many items into one root hash, and a Merkle proof shows one item belongs under it; this page covers Bitcoin and Ethereum uses.

Running a Node & Client Diversity

Running a node means checking a blockchain's rules yourself, and client diversity means no single software bug should decide what the network accepts.

Bitcoin Upgrades: SegWit, Taproot & Soft Forks

Bitcoin upgrades by soft fork: how SegWit and Taproot changed signatures and scripts, how activation worked, and where covenant proposals stand.

When an Exchange Fails: Custodial Claims & Bankruptcy

How courts treat customer balances when a crypto exchange enters bankruptcy, with estate property, dollar claims and three real cases as of October 2026.

Blockchain Forensics: How Crypto Is Traced

Blockchain forensics reads public ledger data, using address clustering, labels and legal process, to follow funds and link them to real parties.

References & further reading