In briefGraphene double quantum dot as a terahertz-range solid-state two-level system
SummaryThe study uses an hBN-encapsulated graphene double quantum dot to form an artificial molecule, with bonding and antibonding states separated by up to 0.6 meV. This corresponds to a resonance frequency of 0.145 THz. Low-temperature transport measurements combined with nonequilibrium Green's function modeling provide direct access to parameters such as tunneling rates. The authors propose the device as a terahertz two-level system compatible with on-chip integration.
Alexis Wietzke, Solen Coeymans, Benjamin Hue et al.
In briefA study of verification flaws in cross-chain bridge smart contracts
SummaryThis paper presents the first systematic empirical study of vulnerabilities in the verification logic of cross-chain bridge contracts, which validate external-chain data such as state proofs and signatures. The authors analyzed 588 security audit reports from 36 real-world bridge projects and identified 11 vulnerability patterns grouped into 5 categories. They built a dataset of 830 vulnerability instances across 786 contract files. Evaluation shows existing tools cover these patterns only in limited ways, and the best of four LLMs reaches 51.2% Macro-F1, motivating detection methods tailored to cross-chain verification.
Yuying Wang (Institute of Computing Technology, Chinese Academy of Sciences, China et al.
In briefSwingGuard: Risk-Conditioned Multi-Agent Crypto Swing Trading with Two-Speed Oversight
SummaryThis paper presents SwingGuard, a long-only multi-agent framework for crypto swing trading that operates on two timescales. A strategy layer has analysts weigh technical, sentiment, on-chain and macro evidence, then a portfolio architect sets a target position conditioned on current risk. A five-minute risk controller monitors risk and constrains execution until the next strategy revision. Tests on Bitcoin and Ether report positive returns after transaction costs and shallow drawdowns, with a joint portfolio Sharpe ratio of 2.98 versus 2.79 for a post-hoc risk variant.
In briefAgentValve: A payment authorization protocol resistant to Sybil attacks in open AI markets
SummaryThis paper presents AgentValve, a protocol that authorizes AI agent payments in permissionless markets while resisting Sybil attacks, where adversaries fabricate identities and feedback to inflate provider reputation. It models reputation as finite capacity in a directed trust graph and uses a smart contract to reserve capacity for each admitted payment. Experiments using six months of Ethereum data and synthetic workloads report that payment exposure and capacity depletion stay bounded with up to 10,000 synthetic Sybil identities. The witness-commit mechanism reduces admission cost by up to 40.8%, with common payments costing about $0.04 to $0.21.
In briefFrom Imbalance to Evidence: A Deep Classifier Test for Very Uneven Sample Sizes
SummaryThis paper studies how to compare two data distributions when one sample is far smaller than the other, a setting where classifier-based two-sample tests lose power. The authors propose EIAD-C2ST, a classifier-agnostic framework that turns the surplus majority-sample observations into useful testing evidence, with a scalable implementation for deep classifiers. The method comes with theoretical guarantees on size and power, and is evaluated through simulations and a blockchain forensics application.
In briefDeFiRank: Federated Pairwise Ranking for Fair Governance in Decentralized Finance
SummaryThis paper proposes DeFiRank, a federated learning framework in which multiple institutions jointly train a pairwise ranking model without sharing private transaction data. It ranks providers by the authenticity of their actions, promoting genuine contributors and demoting suspicious ones. On a real blockchain transaction dataset, the federated method's NDCG@15 is about 0.87 versus about 0.92 for the centralized baseline, with similar Kendall τ (about 0.70) for both.
Zahra Batool, Michael Duchesne, Kaiwen Zhang et al.
In briefMagnetar: a fast post-quantum RV64 zkVM built on standard hash functions
SummaryThis paper presents Magnetar, a transparent proof system that proves executions of RV64I programs (a zkVM). Its security rests on standard hash functions such as SHA-256 or SHA-3, without relying on algebraic hashes like Poseidon, elliptic-curve or lattice assumptions. The authors report experiments at 120 bits of provable security showing higher proving throughput than several compared zkVMs.
WhyPresents the first zkVM combining general RV64I proving, standard-hash post-quantum security, and efficient recursion, with broad benchmark comparisons against prior systems.
In briefEfficient Unbiasable Verifiable Random Functions via Reductions
SummaryThis paper presents simple, practical constructions of verifiable random functions (VRFs) that meet strong unbiasability notions introduced by Giunta and Stewart at Eurocrypt 2024. A compiler turns any VRF into an unbiasable VRF in the standard model, without setup, using only symmetric-key primitives and negligible extra computation. The authors also give a non-black-box reduction with polynomial loss under a mild condition, enabling a provably secure "Hash-then-VRF" construction in place of the heuristic "VUF-then-Hash" approach at almost no cost.
WhyGives a symmetric-key compiler making any VRF unbiasable in the standard model, plus a novel polynomial-loss non-black-box reduction; proofs-focused with no experiments.
In briefDiffusionART: an efficient, stable RL method for fine-tuning diffusion models
SummaryThis paper introduces DiffusionART, an online reinforcement learning method that fine-tunes diffusion models using black-box rewards. It aggregates each prompt's rewards into a shared direction and adds a trust-region-masked second update, aiming for stable training without tuning exponential-moving-average settings. The authors report efficient, stable training across image and video generation, including fewer GPU hours than DiffusionNFT and Flow-GRPO-Fast to reach comparable late-stage mean rewards on HunyuanVideo-13B.
Zhenglin Wan, Jiayu Chen, Zhengfeiyang Zhang et al.
In briefIndividual cryptography based on iterated squaring
SummaryThis paper studies individual cryptography and complete knowledge protocols, which ensure that some individual party must know a secret in the clear so the computation cannot be outsourced to MPC. Using ideas from time-lock puzzles, the authors construct secret sharing with snitching and encryption with self-incriminating proofs where reconstruction or decryption must be sequential. Security relies on a new knowledge assumption.
Georg Fuchsbauer, Christoph U. Günther, Charlotte Hoffmann et al.
In briefOne Node, Two Roles: Competing Rewards for Verification and Attention in Rollups
SummaryThis paper studies incentives in optimistic rollups, where safety requires at least one honest validator to re-execute the state transition and dispute incorrect assertions, yet correct verification goes unpaid. The authors formalize arguments with registered provers and model validation and attention rewards as two coupled Tullock-like contests. They compare the TRACE and Proof of Diligence attention mechanisms and apply the results to Arbitrum One.</summary>
Pranay Anchuri, Ben Berger, Matteo Campanelli et al.
