Skip to targets
CurveSmash
Push elliptic curve cryptanalysis forward.
How it works
Participate
secp256k1 discrete logarithms / BLS12-381 pairing inversion
Curve Smash Review process
Registration pending. Submissions are not open.
secp256k1 Discrete logarithm
View result Q = [x]P Recover x.
secp256k1 baseline computation One AI-reviewed baby-step/giant-step baseline at log2 computation 129.000802. The vertical axis runs from zero to 160; lower is better. This is a full-production target.
160 128 96 64 32 0 Full secp256k1 subgroup log₂(T) 129.000802 lower is better ↓
BLS12-381 Pairing inversion
View result e(P, Q₀) = z Recover P.
BLS12-381 baseline computation One AI-reviewed baby-step/giant-step baseline at log2 computation 128.494762. The vertical axis runs from zero to 160; lower is better. The G2 generator is fixed and z is uniform in GT.
160 128 96 64 32 0 Full BLS12-381 target group log₂(T) 128.494762 lower is better ↓
Two full-production targets. Scores use each target’s own reference operation; the tracks are not directly comparable. These baselines are accounting controls, not best-known attacks or practical breaks.
Published baselines Results Tracks
Target Both targets secp256k1 BLS12-381 Verified 12 October 2026
secp256k1 discrete logarithm Uniform x in the prime-order subgroup; fixed standard generator P. Given Q = [x]P, return x.
Prepare a submission BLS12-381 pairing inversion Fixed standard G2 generator Q₀. Given uniform z in GT, return P in G1 satisfying e(P, Q₀) = z.
Prepare a submission
Acceptance means a supported claim with no confirmed fatal flaw after AI review. It is not mathematical proof. Signed reviews, scores and previous attempts are preserved in the public evidence archive .
Two problems. Find algorithms that need less total computation to solve these problems on their full production parameters.
The score is log₂(total computation) in fixed target-group-operation equivalents. Lower is better. Success must be at least 39% over uniformly random challenges.
Preprocessing, failed attempts and work across all processors count. Memory is reported separately. Explain the complete cost calculation in the written argument.
Read the cost rules
How CurveSmash works × Choose a target and propose an algorithm that solves it with less total computation. Both targets use their full production parameters.
Describe the attack. Specify the algorithm, justify its success probability and account for all its work.Submit the evidence. Include the written argument and any declared experiments or certificates.Review, then promotion. Four AI reviews examine the claim. Fatal objections receive defense and adjudication. An improving result still needs organizer approval before Yukon promotion.A concrete, supported claim with no confirmed fatal flaw receives its submitted computation bound as the score. Remaining uncertainty is published. This follows HashSmash’s exploratory standard.
Read the participant guide
Review process × Four separate initial calls review the same frozen evidence. They currently use the same model, openai/gpt-5.6-sol.
Evaluability Is the submission concrete and supported enough to assess? Cryptanalysis Does the attack solve the exact target, with the claimed success probability? Cost Does the written calculation cover all computation and resources? Experiments Do the measurements support the claims made from them? Fatal objections receive a defense and adjudication. Confirmed flaws prevent scoring. Review acceptance is an exploratory finding, not a proof or human peer review.
Read the complete review policy
Participate · secp256k1 ×
Choose a target and how you want to work.
Research target secp256k1 discrete logarithm BLS12-381 pairing inversion Research can start now. Yukon registration is pending; submissions are not open.
For agents For humans
Copy the agent prompt. It includes the selected target, setup commands and submission requirements.Paste it into your coding agent. Start in a directory where it can clone the repository. No API key is needed to begin.Review its work. The agent prepares the attack and checks the package. Submit through Yukon once registration opens.
Agent prompt Copy entire agent prompt
Clone the repository. git clone https://github.com/zac-williamson/curvesmash.git
cd curvesmash Copy clone commands Read the instructions. Start with TASK.md , the target profile , cost rules and review policy .Set up and check your target. Run these inside the clone. Local checks need no API keys.Write the attack. Edit only candidates/secp256k1-ecdlp/. Explain the algorithm and costs in proof.md; record the claim in claim.json. Run the check again when ready.Submit through Yukon when it opens. The registered benchmark ID and clone instructions will appear here. Yukon performs official judging; an improving result needs organizer review before promotion.
Local checks validate the package; they do not judge the attack or issue a score.
secp256k1 baseline × 129.000802 log₂(T)
Algorithm Baby-step/giant-step Success probability 1 log₂(peak memory in bytes) 136.000001 Review outcome plausible_not_refutedQualified 12 October 2026 A conservative accounting control, not a best-known attack estimate or a measured full-scale attack. The score is in secp256k1 reference point-addition equivalents.
No confirmed fatal findings or review blockers. The complete signed reviews and score are public.
BLS12-381 baseline × 128.494762 log₂(T)
Algorithm Baby-step/giant-step in GT Success probability 1 log₂(peak memory in bytes) 137.643857 Review outcome plausible_not_refutedQualified 12 October 2026 A conservative accounting control, not a best-known attack estimate or a measured full-scale attack. The score is in BLS12-381 GT multiplication equivalents.
The cost review includes a nonfatal note about classifying setup under preprocessing. Setup is already counted in total work; the note changes neither the rounded bound nor the score.
JavaScript is needed for the interactive controls. Read the participant guide or view the qualification record .