EIP-8292 Quantum-resistant proof aggregator (Post-Quantum Attestation Aggregators)

No call decision · Consensus & staking · EIP status Draft · created 2026-06-07

In a consensus scheme that has switched to quantum-computer-safe, hash-based signatures, this defines a new participant role, the aggregator, that folds the votes of many validators into a single succinct proof.

Why it is needed

Today the beacon chain authenticates validator votes with BLS signatures. BLS signatures are small, 96 bytes, and have the property that summing multiple signatures still yields a single 96-byte signature, so an entire committee's votes can fit in one block. Hash-based signatures (an XMSS variant) adopted as a defense against quantum computers lack this property. A single signature is about 1.17KiB, more than ten times larger than a BLS signature, and N signatures require N times the space and verification cost. So instead of combining signatures directly, quantum-resistant consensus aggregates by generating a cryptographic proof that "all of these signatures are valid" using a dedicated virtual machine called leanVM. Generating this proof is far heavier than BLS's simple addition. If every validator were required to generate proofs, commodity hardware could not finish within a slot, leading to centralization or missed slots. This is why proof generation needs to be separated from block production and defined as a role for aggregators who have sufficient hardware and opt in voluntarily.

What changes

Today

After EIP-8292

Who is affected

In depth

Why it cannot aggregate like BLS

BLS signatures are points on the BLS12-381 curve, so summing multiple signatures still yields a single 96-byte signature, and the whole set is verified with a single pairing equation. The signatures used in quantum-resistant consensus, by contrast, are a stateful XMSS variant; in exchange for relying only on hash function security, they lack any structure for cheaply combining signatures. Under the parameters currently used by the leanVM aggregation engine, a single signature is about 1.17KiB, and N signatures cost roughly N times as much. So aggregation shifts from combining signatures to generating a quantum-resistant succinct argument that "all attached signatures are valid." Measured aggregate proof sizes range from 127 to 344KiB depending on configuration, and the networking target is to propagate a roughly 256KiB proof within 1 second, a size that is nearly independent of the number of aggregated votes.

Specification of the aggregator role

A node MAY opt into the aggregator role, for example by enabling an is_aggregator flag in its configuration. Aggregation is not a separately drawn validator duty, and participation MUST NOT be gated by a protocol-defined stake threshold. An aggregator MUST subscribe to at least one subnet, and for each vote message in its assigned subnets it MUST collect signed attestations, prove them as a single aggregate, and propagate it on the aggregation topic. Collection and proving SHOULD finish within the intra-slot aggregation window defined by leanSpec, since an aggregate that arrives too late for the proposer to include has no value. Nodes whose validators' combined stake is at least SUBNET_COMMITTEES * 32 ETH SHOULD subscribe to all subnets and act as super-aggregators. This mirrors the arrangement of PeerDAS supernodes. This role does not add any new consensus rule or fork-choice rule.

The three stages of aggregation and role boundaries

Aggregation proceeds through stages with different cost profiles. Leaf (signature) aggregation is the first stage, proving that the raw signatures attached to a single vote message form a single aggregate; it is relatively cheap for a flat batch. Recursive (proof) aggregation folds multiple partial aggregates of the same message again via leanVM, and its cost grows with the number and depth of aggregates being folded. Block-level proof aggregation folds the per-message aggregates into a single block proof within the MAX_ATTESTATIONS_DATA (8) limit, but this is the proposer's duty, not the aggregator's. This EIP only fixes that the aggregator role covers the leaf and recursive stages. The exact aggregation algorithm, recursion strategy, and intra-slot timing are left to leanSpec.

Target figures and measured resource requirements

Aggregation targets are non-normative soft targets rather than consensus requirements. The hard bound on aggregate proof size is 512KiB (the leanSpec container limit), with typical sizes around 185-230KiB, almost independent of the number of votes. About 185KiB for 125 signatures and about 227KiB for 1000, growing only gradually. Proving time, on the other hand, is hardware-bound and sets the lower limit for this role. Aggregating 1000 signatures takes about 11.0 seconds on 2 cores/15GiB, about 3.7 seconds on 8 cores/59GiB, and about 2.6 seconds on 16 cores/118GiB, with core count mattering more than clock speed for meeting the slot budget. Memory use is about 1.5-6.6GiB depending on batch size and recursion depth. Minimum hardware specs are not yet mandated, and for now aggregators are expected to meet the EIP-7870 full node baseline. Keeping proof size and proving time as soft targets is a safeguard against hardware requirements quietly ratcheting up.

The difference from existing committee aggregators, and the rationale behind the role's design

The beacon chain already has an aggregation duty in which randomly selected committee members combine BLS signatures (see EIP-7549's change to the committee index). That duty is cheap point addition, and selection happens secretly via a signature-based selection proof. The quantum-resistant aggregator is different: it is a computationally intensive task of generating proofs for signatures that cannot be combined directly, and on current devnets it is a statically configured, voluntary role rather than a secret lottery. The reason it isn't made a duty for every validator is to keep heavy proving off the critical path of home staking hardware. Proving is left to wherever the hardware exists while verification stays cheap for everyone, an asymmetric structure with the same motivation as PeerDAS (EIP-7594), which offloads heavy custody work to participants with more resources.

Security considerations

Aggregators sit on the path between signatures and finality, so if aggregators stall or selectively drop votes, proposers may fail to include sufficient coverage and finality degrades. The current requirement that there be at least one aggregator formalizes this, and running several redundant aggregators is the main mitigation. The safety of aggregation reduces to the soundness of the leanVM proof system. If a false proof were possible, an attacker could forge consensus participation; the current leanVM instance is reported at about 124 bits classical and 62 bits quantum security, short of the NIST Level 1 target (128 bits classical, 64 bits quantum), and work is underway to close the gap with larger digests and similar measures. Centralization from proving hardware is the biggest concern, but even a captured or colluding set of aggregators must not be able to undermine the widely performed cheap verification. Relying on implicit incentives alone for expensive proving, with no explicit rewards or penalties, is flagged as risky, and whether to publish a minimum hardware spec remains an open question.

Out-of-scope items and draft status

This EIP defines only the aggregator role and leaves everything else to leanSpec, the quantum-resistant consensus specification. The leanSig/XMSS signature scheme and its parameters, the leanVM instruction set and proof system (WHIR, SuperSpartan, Logup), the SSZ byte layouts for attestations, aggregates, and proofs and the gossip message schema, the procedure for migrating validator keys from BLS to a hash-based scheme, and the consensus and fork-choice rules are deliberately out of scope. The SUBNET_COMMITTEES value is still TBD even within leanSpec. The activation hard fork is specified by the relevant meta EIP, and pre-fork clients cannot verify post-fork aggregation proofs unless they implement the quantum-resistant consensus changes. This EIP is currently in Draft status and may change before the reference implementation settles on devnets.

Original EIP · ← EIPs