EIP-8433 Retirement of 0x00 validators (Retire 0x00 Validators)

No call decision · Consensus & staking · EIP status Draft · created 2026-09-28

Validators whose withdrawal credentials are the legacy 0x00 type are force-exited within a per-epoch cap after a grace period. Balances are untouched, and switching credentials to 0x01 allows the full balance to be reclaimed at any time.

Why it is needed

0x00 is the withdrawal credential format from Ethereum's launch. Withdrawals are controlled by a BLS key and there is no execution-layer address. After the Capella upgrade opened a one-way transition to 0x01, the roughly 600,000 0x00 validators that existed then have fallen to about 9,200 (as of mainnet in August 2026), but the decline has stalled. The number of migrations has dropped to double digits per month, and among the still-active 0x00 validators, several hundred appear to have lost their keys, having not attested in over six months. Yet no one can exit a 0x00 validator that has lost its key. Voluntary exit requires a signing key, and EIP-7002, which triggers exits from the execution layer, requires an execution address that 0x00 does not have. Natural attrition through penalties alone, at the measured penalty rate (about 0.55 ETH per year), would take roughly 28 years. In the meantime every fork must keep maintaining 0x00-related code, and when the transition to an era where BLS signatures are broken by quantum computers arrives, these validators, which cannot act, become a burden on the whole design.

What changes

Today

After EIP-8433

Who is affected

In depth

Two constants set the schedule and rate

RETIREMENT_START_EPOCH is the first epoch at which retirement begins, and MAX_RETIREMENTS_PER_EPOCH is the maximum number of exits that can be initiated at each epoch transition; both are TBD. RETIREMENT_START_EPOCH is an absolute epoch, not an offset counted from fork activation. This keeps the grace period unchanged even if the fork schedule shifts, allowing the date to be announced in advance, and lets the date be adjusted during review without using the grace period as a separate fork.

process_bls_validator_retirement epoch processing

A new step, process_bls_validator_retirement, is added to process_epoch after process_registry_updates. If the current epoch is at or past RETIREMENT_START_EPOCH, validators that are active, whose withdrawal_credentials first byte is BLS_WITHDRAWAL_PREFIX, and whose exit_epoch is FAR_FUTURE_EPOCH are found in index order and initiate_validator_exit is called on them. Processing stops once MAX_RETIREMENTS_PER_EPOCH is reached for that epoch. This is not a one-time migration event but a standing rule, so validators that later rise through the activation queue with 0x00 credentials are retired under the same rule. Exits go through the standard exit churn as usual.

Operations that do not change

process_bls_to_execution_change remains unchanged. When a retired 0x00 validator switches to 0x01, once withdrawable_epoch has passed, the withdrawal sweep moves the full balance to the execution address given in the message. process_voluntary_exit also remains unchanged, and a voluntary exit before RETIREMENT_START_EPOCH has the same effect as retirement. Deposit processing is also unchanged by this EIP. Under EIP-8365, deposits that would create a new 0x00 validator are already rejected, and additional deposits to existing validators, including retired 0x00 validators, continue to be credited.

Why exit, and why a standing rule

Exit is the existing mechanism that removes a validator from all duty selection, including proposer, attester, and sync committee, while simultaneously ending both reward accrual and penalty exposure. An approach that removes validators from duties only, without exiting them, was rejected because it would create a new validator state that every committee selection, reward path, and effective-balance calculation would have to handle separately. Exiting everyone at once at the start epoch was also rejected, because it would miss validators sitting in the deposit/activation pipeline that rise a few weeks later, requiring a second special-case event. A per-epoch rule faithfully expresses the invariant that 0x00 is not a valid credential for an active validator, and handles the existing population and late joiners through the same code path.

The cap's tradeoff

The value of MAX_RETIREMENTS_PER_EPOCH is a choice between drain duration and churn-queue share, and remains TBD. For a population of about 9,000, a value of 8 matches the exit churn throughput (MAX_PER_EPOCH_ACTIVATION_EXIT_CHURN_LIMIT / MIN_ACTIVATION_BALANCE = 256 ETH / 32 ETH) exactly, finishing in about 5 days without exceeding the queue's drain rate, while a value of 1 uses 12.5% of churn capacity and takes about 6 weeks. Whichever value is chosen, the time for the entire population to pass through the exit queue does not drop below the roughly 5 days set by churn. Per-epoch cap processing is a pattern already used for pending deposits and pending consolidations.

The logic of credible neutrality, and precedent

Earlier proposals to rescue 0x00 validators that lost their keys were rejected because they would require adjudicating off-chain ownership claims and picking beneficiaries. This EIP takes the opposite shape. It applies a uniform, pre-announced rule to the entire set defined by an objective on-chain property, keeps open to the end a remediation path anyone can use, BLSToExecutionChange, and takes no position on any ownership claim. Pre-announced, rule-based deprecation that works against specific users is already an accepted category, in the same shape as opcode repricing (EIP-2929), SELFDESTRUCT's disabling (EIP-6780), and inactivity leaks. The SELFDESTRUCT deprecation itself also followed a staged path across multiple EIPs and forks, as with EIP-6049's advance notice and EIP-6780's behavioral restriction.

Security implications: stake, ownership, and the exit queue

Even if the entire current population were retired, at most about 300,000 ETH of effective balance would leave the active set, roughly 1% of total stake at the time of writing, and this share only shrinks as the transition proceeds. This falls within the normal range of validator churn, so it has no meaningful effect on finality safety margins or the weak subjectivity period. No balances move at all. Every retired validator's balance remains in the beacon state and can be withdrawn in full via the exit path, and holders who have already lost their withdrawal keys and cannot use that path are no worse off than they already are. Thanks to the churn cap, the exit queue does not build up a backlog, and even if demand temporarily exceeds throughput because it overlaps with ordinary voluntary exits in the same period, the worst case is a mild wait comparable to a busy exit period.

Second stage of a four-stage deprecation, still Draft

This EIP is the second stage of deprecating 0x00 credentials. EIP-8365 blocks inflow by rejecting deposits that create new 0x00 validators, and this EIP removes the existing population from duty participation. A follow-up companion balance sunset EIP then gradually reduces retired 0x00 balances to zero ahead of the quantum-resistant transition, and a final-stage EIP removes the remaining 0x00 machinery from the consensus layer at the quantum-resistant fork, including BLSToExecutionChange, its gossip topic, the operations pool, and zero-balance registry entries. The stages are designed to activate across consecutive forks, serving as a timetable holders can prepare against. The source is still Draft, so the content, including the values of both constants, may still change, and test cases are TBD.

Original EIP · ← EIPs