# Enabling Threshold Custody for the Lightning Network with Nested Threshold Multi-Signatures

Crypto.world

> Every Lightning channel endpoint is protected by a single online key, so a compromise or loss of that one key loses the channel's funds — and unlike on-chain Bitcoin, cold storage is not an option…

**arXiv**: 2608.20705

**Published**: 2026-08-21

**Authors**: Paul Gerhart, Nadav Kohen, Jesse Posner, Matias Furszyfer

## AI summary

## Overview

**Research area:** Bitcoin / Lightning Network cryptography — threshold signature schemes, MuSig2 multi-signatures, and Schnorr signature security.

**Technical level:** Advanced. The paper combines a formal cryptographic primitive definition, a provable-security treatment (ROM and ROM+AGM), and systems benchmarking, and it assumes familiarity with Schnorr signatures, nonce commitments, and Lightning channel mechanics.

**Scope:** The paper formalizes "nested threshold multi-signatures," presents and proves the security of *Iceberg* (the first nested threshold MuSig2 construction), and benchmarks a prototype inside a production Lightning node to show that one channel endpoint can be operated as a *t*-of-*n* group without any change to Bitcoin, the Lightning protocol, or the counterparty.

## What This Paper Is About

Every Lightning channel endpoint is protected by a single online key, so a compromise or loss of that one key loses the channel's funds — and unlike on-chain Bitcoin, cold storage is not an option because the key must stay online to route payments. The goal is to let one side of a channel be split into a *t*-of-*n* threshold group while the counterparty and the blockchain continue to see an ordinary single MuSig2 signer, deployable unilaterally on today's network. The obstacle is that thresholdizing one participant requires nesting a threshold scheme inside an unmodified two-party MuSig2 session, whose nonce is fixed a full round-trip before the message to be signed exists.

## Key Contributions

1. **A new primitive: nested threshold multi-signatures.** The authors formalize the notion of *t*-of-*n* threshold-signing the partial signatures of an existing multi-signature scheme, covering the case where a threshold group occupies exactly one slot of an outer multi-signature (Section 4).
2. **Iceberg, the first nested threshold MuSig2 construction.** Iceberg lets one side of a Lightning channel act as a *t*-of-*n* group while appearing to the counterparty as a standard MuSig2 participant (Section 5).
3. **A security proof with Lightning-appropriate thresholds.** Unforgeability holds when an adversary compromises up to *t*−1 signing key shares, and any 2*t*−1 signing shares suffice to produce a valid signature (Section 6).
4. **Compatibility and a measured prototype.** Iceberg's first-round output is exactly the message a single MuSig2 participant would send and the funded channel is an ordinary single-key Taproot output, so it is a drop-in MuSig2 replacement; the authors integrate it into a production Lightning node and benchmark it over taproot channels (Section 7).

## Main Findings

- **Threshold custody is deployable today without protocol changes.** An Iceberg endpoint requires no modification to Bitcoin, the Lightning protocol, or the channel counterparty; adoption is unilateral.
- **Measured overhead of a 2-of-4 group:** making one endpoint a 2-of-4 group adds **3.8 ms per payment**, which is **6.7%** of the work a payment already costs, and that group only saturates an additional core at **260 payments per second**.
- **Throughput retained:** a threshold group tolerating one corrupted member sustains over **93%** of the payment throughput of an unmodified endpoint.
- **Signing parameters are dictated by the channel, not chosen freely.** Because the group must agree on the live channel state — a Byzantine agreement problem — the construction requires an honest supermajority, hence **n ≥ 3t − 2**, a signing quorum of **2t − 1** online members, and tolerance of **t − 1** corruptions. Concrete table values: *t*=2 tolerates 1 corruption, needs 3 online signers, and requires a group of at least 4; *t*=3 → 2 corruptions, 5 signers, group of 7; *t*=4 → 3, 7, 10; *t*=5 → 4, 9, 13.
- **Deterministic nonces are made safe by the channel state, not the message.** Nonces are derived from a session identifier that includes the commitment number (unique and strictly increasing per channel), so the nonce is fixed before the message exists yet varies per channel state. The residual risk — two messages signed under one session identifier — is excluded by the group's pre-existing need to agree on the live commitment number, since signing a revoked state triggers Lightning's penalty mechanism.
- **Replicated secret sharing solves quorum changes.** Deriving nonces deterministically from replicated shares means any qualifying quorum recomputes the same group nonce, and a member that was offline in the first round can still sign in the second — unlike FROST, whose nonce depends on the signing quorum.
- **Storage cost is acceptable only because signer sets are small.** Each member holds binomial(n−1, t−1) shares, out of binomial(n, t−1) key summands in total; the authors expect *n* to be at most a dozen.
- **Proof strategy:** Iceberg's *t*-of-*n* messages are expanded into a virtual binomial(n, t−1)-out-of-binomial(n, t−1) multi-signature, "Iceberg_base" (Lemma 1), which is then reduced to NestedMuSig2; because NestedMuSig2 has proofs in the ROM (using ν=8 at depth 2) and in the ROM + Algebraic Group Model (using ν=2, as in BIP 327), Iceberg inherits both bounds.

## Methodology in Plain English

The authors first list the constraints the Lightning environment imposes on any thresholdized endpoint: signers go offline routinely, the group's nonce must be committed before the transaction to be signed exists, that nonce cannot later be changed (the outer session is unmodified BIP 327), the group must agree on the current channel state or risk the penalty, and signer sets are small.

They then show why off-the-shelf tools fail. Nesting plain MuSig2 gives only an *n*-of-*n* group where everyone must be online. FROST and similar Shamir-based schemes build the aggregate nonce from whoever signs, so a different finishing quorum would produce a different nonce than the one already promised to the counterparty. Fully stateless schemes such as Arctic derive the nonce from the message, which is unknown when Lightning requires the nonce, and statelessness would also let signers be tricked into signing a revoked state.

Iceberg combines three design choices: (1) replicated secret sharing, so reconstruction doesn't depend on who is present; (2) deterministic pseudorandom nonce derivation seeded with a session identifier containing the channel's commitment number, evaluated per share; and (3) use of the group's existing consensus on the live channel state as the safeguard against nonce reuse. For security, they extend Kohen's nested-MuSig2 model (adding an inner corruption threshold, quorum changes between rounds, and state agreement), translate the threshold protocol into an equivalent large multi-signature, and reduce that to NestedMuSig2 to inherit its bounds. Finally they implement the scheme in a production Lightning node and measure payment latency and throughput over taproot channels.

## Why This Matters

**Impact on research.** The paper defines a primitive that sits between threshold signatures and multi-signatures — a threshold group occupying a single slot of an outer multi-signature — and supplies the first construction, proof, and deployment measurement for it. It also shows how replicated secret sharing and pseudorandom secret sharing, usually dismissed as impractical for large deployments because of share-count growth, become practical in small-signer settings.

**Real-world applications:**

- Exchanges that settle customer deposits and withdrawals over Lightning (Bitfinex, OKX, Kraken, Binance, and Coinbase are named as doing so) can distribute a channel's hot key across servers and staff instead of holding one online key.
- Self-custodians and institutions can spread a routing node's key across devices held in different locations, including cold-storage-like arrangements, without losing the ability to keep the channel online.
- Lightning infrastructure and routing-node operators gain survivability: losing up to *n* − *t* shares is recoverable and compromising up to *t* − 1 shares does not allow forgery.
- Wallet and node implementers can offer threshold custody as a unilateral feature, since no counterparty coordination or script changes are needed.

**Industry relevance.** The motivating numbers are concrete: roughly 5,600 BTC (about $490 million) in public channels as of December 2025, on the order of 15,000 public nodes with figures described as lower bounds, plus the recurring history of key compromise (7,000 BTC drained from a single exchange hot wallet in 2019) and permanent loss. Threshold custody is already standard on-chain, and this work extends it to the layer where the single online key is unavoidable.

## Future Directions

- **Trusted dealer versus distributed key generation.** The paper states the shares can be set up by a trusted dealer or by a distributed key generation, but the deployment and measurement details of each are not reported in the available content.
- **Larger signer sets and share growth.** Since per-member storage is binomial(n−1, t−1) shares and the authors expect *n* at most a dozen, the boundary at which replicated sharing stops being practical, and alternatives for it, remain open.
- **Failure and retry handling at the outer MuSig2 layer.** The paper notes that if an implementation wants to retry a failed signing attempt, it must agree on protocol messages that change the session identifier; the concrete mechanism is left to implementations.
- **Aggregator trust and privacy.** The group uses an untrusted aggregator node, which may be one of the participants or another party — and using another party leaks privacy, a trade-off that invites further design work.

## Target Audience

Cryptography and Bitcoin-protocol researchers interested in threshold and multi-signature schemes; Lightning node, wallet, and custody implementers who need deployable key-distribution options; and security architects at exchanges and custodians who must justify why a single online channel key is, or is not, acceptable. Readers without a background in Schnorr signatures, nonce commitments, and secret sharing will find the security sections and formal preliminaries demanding.

## Authors’ abstract

The Bitcoin Lightning Network secures hundreds of millions of dollars, yet channel endpoints rely on vulnerable single online keys. Although threshold signatures are routinely used to protect on-chain Bitcoin, no practical deployment has been possible for Lightning channels. This is because thresholdizing a Lightning party requires nesting a threshold signature scheme inside of an established two-party MuSig2 protocol without altering its nonce exchange or message flow. In…

[The full abstract on arXiv](https://arxiv.org/abs/2608.20705)

[Read the original paper](https://arxiv.org/abs/2608.20705)

---

Canonical: https://crypto.world/paper/enabling-threshold-custody-for-the-lightning-network-with-nested-2608.20705
