SHRINCS, the hash-based signature scheme that a group of Bitcoin developers is drafting as a defence against quantum computers, now has a written security proof. Mikhail Kudinov, a cryptographer at Blockstream, put a 131-page analysis dated 18 November 2026 in a public GitHub repository and announced it on X the next day, Thursday 19 November. It concludes that the scheme holds about 121 to 123 bits of security against classical attackers, a few bits under its 128-bit target, and it says so under three conjectures that nobody has proved.
The paper fills a gap in the SHRINCS draft: when the draft was first posted to the bitcoin-dev mailing list on 26 August, a security proof was on its list of missing parts. It decides nothing about Bitcoin itself. No soft fork is proposed, no date exists, and the draft has no BIP number. From its second page on, every page of the paper carries the same footer, which calls it a “Proposal for discussion”, “not a specification”, and adds: “Not audited; not for deployment.”
Kudinov made the same point in his thread. “We tried to be careful, but a proof of this size can have mistakes,” he wrote on 19 November, asking readers who find one to open an issue in the repository. The second post of the thread says the work “hasn’t been sufficiently peer reviewed yet”.
The paper landed in a busy week for the subject. On 19 November a census counted 7.16 million BTC sitting behind public keys already visible on-chain, and the same day IOSCO warned that blockchains face a hard post-quantum upgrade. As late as 12 May, Nick’s post on the Blockstream blog opened by saying there was no concrete proposal for a post-quantum signature scheme in Bitcoin. The SHRINCS draft followed in August.
What the SHRINCS security analysis proves
The PDF and the repository README name one author, Kudinov. Jonas Nick, also of Blockstream, is a co-author of the draft specification and wrote the earlier blog post that proposed a SHRINCS opcode; the two documents should not be confused. The analysis restates the full scheme, then proves (Theorem 4.1) that forging a SHRINCS signature is as hard as breaking specific properties of the hash function underneath. That function is SHA-256, the one Bitcoin already uses for blockchain and earns a reward, which makes rewriting history very expensive.">mining and transaction identifiers.
A proof of this kind does not say “unbreakable”. It gives a number: how much work an attacker needs before the guarantee runs out. The draft specification targets what NIST calls security category 1, which it words as “128 bits of classical and 64 bits of quantum security”. Section 6 of the paper computes what the proof can certify for the proposed parameters.
| Part of the scheme | Classical security certified | Quantum hash queries at which the bound runs out |
|---|---|---|
| Stateless path | 122.23 bits | about 2^58.5 |
| Stateful path | 122.75 bits | about 2^59.2 |
| Full scheme | 121.47 bits | about 2^58.0 |
| Target in the draft specification | 128 bits | 64 bits of quantum security |
Source: Table 5 of the security analysis, draft of 18 November 2026, and the draft specification. The paper rounds the classical range to “between 121.4 and 122.7 bits”.
The certified figures sit below the targets. On the classical side the gap is 5.25 bits for the stateful path and 6.53 bits for the full scheme (128 minus the table’s figures). The paper describes no attack at those levels. It presents the gap as a loss of the proof: the “loss against the 128-bit hash”, it says, is the sum of the coefficients of the terms in its bounds. A 121-bit bound still corresponds to about 2^121 hash computations by the attacker.
The author also published two companion papers in the same repository. Kudinov wrote on X that their results are not specific to SHRINCS and also apply to SPHINCS+ and SLH-DSA, the stateless hash-based scheme standardised by NIST.

Three conjectures and four open items
“The figures rest on three conjectures,” the paper states. The first concerns a hash property abbreviated SM-DSPR. Its bound has been conjectured rather than proven in the SPHINCS+ literature, and it enters the figures for the stateless path and for the full scheme. The second (Conjecture 4.1) concerns the way the paper accounts for the work done inside the proof itself. The third (Conjecture 3.1) assumes that a counter search performed by the stateful signer always succeeds within its 2^16 attempts.
All the bounds also treat the hash function as an ideal random function, the usual simplification in this field. The README adds what the analysis does not cover at all: side channels, implementation bugs, key management and misuse of the signing state.
Section 7 lists four results the author expects to improve, among them removing the SM-DSPR conjecture and tightening the quantum bounds. On the first, the paper says the term is a small share of the total, so the numbers should change little. Those drafts are not part of the document, and the paper says they have not been checked to the same standard.
One public key, two ways to sign
Today a Bitcoin signature relies on elliptic curves, which a large enough quantum computer could break with Shor’s algorithm. A hash-based signature relies only on a hash function being hard to reverse. In its simplest form the signer publishes the hashes of secret values and signs by revealing some of those secrets. Each such key can sign safely once. To sign more than once, many one-time keys are gathered in a tree under a single public key.
That is where “state” comes in. A stateful signer keeps a counter of which one-time keys it has already used. The signatures are small, but the counter must never go backwards. A stateless signer needs no counter and pays for it with much larger signatures.
SHRINCS (the name stands for “Shrunken SPHINCS”) puts both under one 48-byte public key. In normal use the wallet signs through the compact stateful path. If the counter is lost, corrupted or simply uncertain, it switches to a stateless fallback built on SLH-DSA with non-standard parameters. The draft specification gives the sizes.
| Signature | Size in bytes | Compared with Schnorr |
|---|---|---|
| Schnorr (BIP 340), used by Taproot today | 64 | 1x |
| SHRINCS stateful, smallest | 548 | about 8.6x |
| SHRINCS stateful, balanced tree of depth 20 | 854 | about 13x |
| SHRINCS stateful, largest | 4,619 | about 72x |
| SHRINCS stateless fallback | 5,777 | about 90x |
| SLH-DSA-SHA2-128s (NIST standard) | 7,856 | about 123x |
Sources: draft SHRINCS specification and security analysis; ratios computed from the byte sizes.
With the “unbalanced” tree shape, the first signature is 548 bytes and each of the next seven from the same key adds 16 bytes. The draft’s reasoning is that a Bitcoin key “is typically used only a few times”. The stateless fallback is 26% smaller than the NIST parameter set because the authors cut the number of signatures a key may ever produce from 2^64 to 2^40, about 1.1 trillion. The draft estimates that 200 years of full 4 MB blocks would use roughly 0.66% of that budget.
The specification also says the smallest public key plus stateful signature is about 13.23 times smaller than SLH-DSA-SHA2-128s and 6.26 times smaller than ML-DSA-44, the lattice-based NIST standard. The second comparison comes with a caveat written in the draft itself: ML-DSA-44 targets NIST category 2, SHRINCS category 1. The two are not at the same security level.
The wallet backup problem
A seed phrase written on paper restores keys. It cannot restore a counter. A wallet rebuilt from its backup has no way to know how many stateful signatures the old device produced, and if it starts again from zero it signs twice with the same one-time key. The draft is blunt about the result: reusing a counter value “can enable anyone who observes the resulting signatures to produce forgeries”. On Bitcoin, a forged signature is what spends the coins a key protects.
SHRINCS answers with a rule. An implementation “MUST refuse stateful signing” whenever the counter is lost, corrupted or uncertain, for instance after restoring from a static backup, and must use the stateless path from then on. The funds stay spendable, with a signature of 5,777 bytes instead of 548 at best, about ten times larger by simple division.
Applying the rule is left to wallet software. The draft’s reference code “performs no state management at all”, and the README of the security analysis says its bounds say nothing about “misuse of the state”. The proof covers a signer that never repeats a counter. In his May post, Nick called a stateful scheme “inherently more fragile than a stateless one” and described one setup in which the user cannot corrupt the counter by accident: a dedicated signing device that creates the state itself and never lets it out.

Why larger signatures matter for fees and for Lightning
Bitcoin blocks have a fixed capacity and users bid for it by the byte, so a signature nine to ninety times larger takes that much more room and fewer transactions fit in each block. The draft sets no price on this. The figures below are estimates by named developers on the bitcoin-dev mailing list, not measurements.
In a message sent late on 31 August (UTC), Lightning developer Antoine Riard noted that the draft says nothing on how signatures would be accounted for. Off-chain protocols such as Lightning and vaults may want to copy their pre-signed transactions to many watchtowers without synchronising a counter between them, he wrote, which points them to the 5,777-byte stateless signature. His estimate: “a 90x-ish of the operational cost for the fee-bumping reserves all those protocols might have to keep”.
Conduition, the pseudonymous developer who posted the draft for the SHRINCS Working Group, answered on 4 September that signature size does not translate directly into fees, because the rest of the transaction counts too. His example is a transaction with two inputs and two outputs: about 840 weight units with Taproot and Schnorr, about 12,426 with stateless SHRINCS in a pay-to-merkle-root output. That is roughly 14.8 times heavier, not 90. He did not put a number on the rest. If SHRINCS were in common use without a bigger block or a discount, he wrote, it is reasonable to assume that throughput would fall and the fee market would inflate: “I can’t conjecture on how high fee rates might escalate in this scenario.”
Nick’s post of 12 May 2026 on the Blockstream blog frames the same issue as “throughput collapse”. By his figures, which assume an average transaction of 2.27 inputs and 2.64 outputs and use the parameters of his proposal at the time, Bitcoin handles about 6.5 transactions per second with Schnorr. That would fall to 0.36 with standard SLH-DSA and to about half a transaction per second with ML-DSA, and it would reach up to about 3 if every signature used the compact SHRINCS path.
One mitigating factor is in the draft. Per byte, SHRINCS signatures cost less to verify than Schnorr ones, which the authors say “leaves room for a witness discount”. The draft then adds: “This specification does not define such a discount.”
Nothing is activated, numbered or scheduled
The SHRINCS document is a draft. Its header reads “BIP: ?” and “Status: Draft”. A search of the bitcoin/bips repository on GitHub on Saturday 21 November at 12:19 UTC returned no pull request mentioning SHRINCS. The mailing-list thread ends on 19 September with conduition writing that the group wants time to review public feedback before submitting one. The draft’s own abstract, as it stood on 21 November, still says “A security proof is TODO”, and a later section says a fully optimised implementation is still to be written.
It is also cryptography only. Conduition compared its role to BIP 340, which defines Schnorr signatures and nothing else. For SHRINCS to be used on Bitcoin, he wrote on 4 September, a further BIP would have to deploy it into consensus and handle the accounting of size and computing cost. A change of consensus rules of that kind is made through a soft fork. Our guide to how Bitcoin Improvement Proposals work and who decides describes that path. None of those steps has been proposed with a date.
BIP 360 is a separate draft, assigned its number in December 2024. It proposes, through a soft fork, an output type called pay-to-merkle-root that works like Taproot with the key path removed. Its own text says this protects only against “long exposure attacks” on keys that sit visible on-chain, and that guarding a transaction while it waits in the mempool “may require the introduction of post-quantum signatures”, left to a later proposal. Nick wrote in May that a hash-based opcode could ship through Taproot, a “Taproot v2” or BIP 360. So BIP 360 is a possible container and SHRINCS a possible content. Neither is adopted. Nick’s post presents his own opcode design under the heading “A Toy Proposal”. His post also states that SHRINCS verification has been demonstrated on Liquid, Blockstream’s sidechain, which is not Bitcoin’s main chain.
Other candidates appear in the same thread: Riard mentioned Falcon as more compact, and conduition listed Falcon, ML-DSA and SQIsign as algorithms that might be available by then. The debate about how urgent all this is runs in parallel: see the Alice & Bob estimate of the machine needed to crack Bitcoin’s curve and the industry alliance trying to define “quantum-resistant”.
What to watch next
The first test is review. The repository asks for errors in proofs, bounds or parameters to be filed as issues, and for citations to name a commit, since the documents will be revised; the commit history shows the PDF replaced once, on 19 November at 09:29 UTC, for “wording and clarifications”. On Saturday 21 November at 12:18 UTC the repository listed no issue, open or closed.
Three documents are still to come. Kudinov says a proof of the SM-DSPR bound is being written. The working group has yet to send its pull request to the BIPs repository. And nobody has posted the consensus proposal that would say how much block space a SHRINCS signature is charged for; the fee figures above are estimates by Riard, conduition and Nick.
